2 * Copyright (c) 2009-2010 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>
52 static char *states[] = {
69 module_param(nocong, int, 0644);
70 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
72 static int enable_ecn;
73 module_param(enable_ecn, int, 0644);
74 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
76 static int dack_mode = 1;
77 module_param(dack_mode, int, 0644);
78 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
80 int c4iw_max_read_depth = 8;
81 module_param(c4iw_max_read_depth, int, 0644);
82 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
84 static int enable_tcp_timestamps;
85 module_param(enable_tcp_timestamps, int, 0644);
86 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
88 static int enable_tcp_sack;
89 module_param(enable_tcp_sack, int, 0644);
90 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
92 static int enable_tcp_window_scaling = 1;
93 module_param(enable_tcp_window_scaling, int, 0644);
94 MODULE_PARM_DESC(enable_tcp_window_scaling,
95 "Enable tcp window scaling (default=1)");
98 module_param(c4iw_debug, int, 0644);
99 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
101 static int peer2peer;
102 module_param(peer2peer, int, 0644);
103 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
105 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
106 module_param(p2p_type, int, 0644);
107 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
108 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
110 static int ep_timeout_secs = 60;
111 module_param(ep_timeout_secs, int, 0644);
112 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
113 "in seconds (default=60)");
115 static int mpa_rev = 1;
116 module_param(mpa_rev, int, 0644);
117 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
118 "1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
119 " compliant (default=1)");
121 static int markers_enabled;
122 module_param(markers_enabled, int, 0644);
123 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
125 static int crc_enabled = 1;
126 module_param(crc_enabled, int, 0644);
127 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
129 static int rcv_win = 256 * 1024;
130 module_param(rcv_win, int, 0644);
131 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
133 static int snd_win = 128 * 1024;
134 module_param(snd_win, int, 0644);
135 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
137 static struct workqueue_struct *workq;
139 static struct sk_buff_head rxq;
141 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
142 static void ep_timeout(unsigned long arg);
143 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
145 static LIST_HEAD(timeout_list);
146 static spinlock_t timeout_lock;
148 static void deref_qp(struct c4iw_ep *ep)
150 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
151 clear_bit(QP_REFERENCED, &ep->com.flags);
154 static void ref_qp(struct c4iw_ep *ep)
156 set_bit(QP_REFERENCED, &ep->com.flags);
157 c4iw_qp_add_ref(&ep->com.qp->ibqp);
160 static void start_ep_timer(struct c4iw_ep *ep)
162 PDBG("%s ep %p\n", __func__, ep);
163 if (timer_pending(&ep->timer)) {
164 pr_err("%s timer already started! ep %p\n",
168 clear_bit(TIMEOUT, &ep->com.flags);
169 c4iw_get_ep(&ep->com);
170 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
171 ep->timer.data = (unsigned long)ep;
172 ep->timer.function = ep_timeout;
173 add_timer(&ep->timer);
176 static void stop_ep_timer(struct c4iw_ep *ep)
178 PDBG("%s ep %p stopping\n", __func__, ep);
179 del_timer_sync(&ep->timer);
180 if (!test_and_set_bit(TIMEOUT, &ep->com.flags))
181 c4iw_put_ep(&ep->com);
184 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
185 struct l2t_entry *l2e)
189 if (c4iw_fatal_error(rdev)) {
191 PDBG("%s - device in error state - dropping\n", __func__);
194 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
197 return error < 0 ? error : 0;
200 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
204 if (c4iw_fatal_error(rdev)) {
206 PDBG("%s - device in error state - dropping\n", __func__);
209 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
212 return error < 0 ? error : 0;
215 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
217 struct cpl_tid_release *req;
219 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
222 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
223 INIT_TP_WR(req, hwtid);
224 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
225 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
226 c4iw_ofld_send(rdev, skb);
230 static void set_emss(struct c4iw_ep *ep, u16 opt)
232 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
234 if (GET_TCPOPT_TSTAMP(opt))
238 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
242 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
244 enum c4iw_ep_state state;
246 mutex_lock(&epc->mutex);
248 mutex_unlock(&epc->mutex);
252 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
257 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
259 mutex_lock(&epc->mutex);
260 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
261 __state_set(epc, new);
262 mutex_unlock(&epc->mutex);
266 static void *alloc_ep(int size, gfp_t gfp)
268 struct c4iw_ep_common *epc;
270 epc = kzalloc(size, gfp);
272 kref_init(&epc->kref);
273 mutex_init(&epc->mutex);
274 c4iw_init_wr_wait(&epc->wr_wait);
276 PDBG("%s alloc ep %p\n", __func__, epc);
280 void _c4iw_free_ep(struct kref *kref)
284 ep = container_of(kref, struct c4iw_ep, com.kref);
285 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
286 if (test_bit(QP_REFERENCED, &ep->com.flags))
288 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
289 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
290 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
291 dst_release(ep->dst);
292 cxgb4_l2t_release(ep->l2t);
297 static void release_ep_resources(struct c4iw_ep *ep)
299 set_bit(RELEASE_RESOURCES, &ep->com.flags);
300 c4iw_put_ep(&ep->com);
303 static int status2errno(int status)
308 case CPL_ERR_CONN_RESET:
310 case CPL_ERR_ARP_MISS:
311 return -EHOSTUNREACH;
312 case CPL_ERR_CONN_TIMEDOUT:
314 case CPL_ERR_TCAM_FULL:
316 case CPL_ERR_CONN_EXIST:
324 * Try and reuse skbs already allocated...
326 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
328 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
331 skb_reset_transport_header(skb);
333 skb = alloc_skb(len, gfp);
338 static struct net_device *get_real_dev(struct net_device *egress_dev)
340 struct net_device *phys_dev = egress_dev;
341 if (egress_dev->priv_flags & IFF_802_1Q_VLAN)
342 phys_dev = vlan_dev_real_dev(egress_dev);
346 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
350 egress_dev = get_real_dev(egress_dev);
351 for (i = 0; i < dev->rdev.lldi.nports; i++)
352 if (dev->rdev.lldi.ports[i] == egress_dev)
357 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
358 __u8 *peer_ip, __be16 local_port,
359 __be16 peer_port, u8 tos,
362 struct dst_entry *dst = NULL;
364 if (IS_ENABLED(CONFIG_IPV6)) {
367 memset(&fl6, 0, sizeof(fl6));
368 memcpy(&fl6.daddr, peer_ip, 16);
369 memcpy(&fl6.saddr, local_ip, 16);
370 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
371 fl6.flowi6_oif = sin6_scope_id;
372 dst = ip6_route_output(&init_net, NULL, &fl6);
375 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
376 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
386 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
387 __be32 peer_ip, __be16 local_port,
388 __be16 peer_port, u8 tos)
394 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
395 peer_port, local_port, IPPROTO_TCP,
399 n = dst_neigh_lookup(&rt->dst, &peer_ip);
402 if (!our_interface(dev, n->dev)) {
403 dst_release(&rt->dst);
410 static void arp_failure_discard(void *handle, struct sk_buff *skb)
412 PDBG("%s c4iw_dev %p\n", __func__, handle);
417 * Handle an ARP failure for an active open.
419 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
421 printk(KERN_ERR MOD "ARP failure duing connect\n");
426 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
429 static void abort_arp_failure(void *handle, struct sk_buff *skb)
431 struct c4iw_rdev *rdev = handle;
432 struct cpl_abort_req *req = cplhdr(skb);
434 PDBG("%s rdev %p\n", __func__, rdev);
435 req->cmd = CPL_ABORT_NO_RST;
436 c4iw_ofld_send(rdev, skb);
439 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
441 unsigned int flowclen = 80;
442 struct fw_flowc_wr *flowc;
445 skb = get_skb(skb, flowclen, GFP_KERNEL);
446 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
448 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
449 FW_FLOWC_WR_NPARAMS(8));
450 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
451 16)) | FW_WR_FLOWID(ep->hwtid));
453 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
454 flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
455 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
456 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
457 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
458 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
459 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
460 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
461 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
462 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
463 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
464 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
465 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
466 flowc->mnemval[6].val = cpu_to_be32(snd_win);
467 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
468 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
469 /* Pad WR to 16 byte boundary */
470 flowc->mnemval[8].mnemonic = 0;
471 flowc->mnemval[8].val = 0;
472 for (i = 0; i < 9; i++) {
473 flowc->mnemval[i].r4[0] = 0;
474 flowc->mnemval[i].r4[1] = 0;
475 flowc->mnemval[i].r4[2] = 0;
478 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
479 c4iw_ofld_send(&ep->com.dev->rdev, skb);
482 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
484 struct cpl_close_con_req *req;
486 int wrlen = roundup(sizeof *req, 16);
488 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
489 skb = get_skb(NULL, wrlen, gfp);
491 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
494 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
495 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
496 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
497 memset(req, 0, wrlen);
498 INIT_TP_WR(req, ep->hwtid);
499 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
501 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
504 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
506 struct cpl_abort_req *req;
507 int wrlen = roundup(sizeof *req, 16);
509 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
510 skb = get_skb(skb, wrlen, gfp);
512 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
516 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
517 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
518 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
519 memset(req, 0, wrlen);
520 INIT_TP_WR(req, ep->hwtid);
521 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
522 req->cmd = CPL_ABORT_SEND_RST;
523 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
526 #define VLAN_NONE 0xfff
527 #define FILTER_SEL_VLAN_NONE 0xffff
528 #define FILTER_SEL_WIDTH_P_FC (3+1) /* port uses 3 bits, FCoE one bit */
529 #define FILTER_SEL_WIDTH_VIN_P_FC \
530 (6 + 7 + FILTER_SEL_WIDTH_P_FC) /* 6 bits are unused, VF uses 7 bits*/
531 #define FILTER_SEL_WIDTH_TAG_P_FC \
532 (3 + FILTER_SEL_WIDTH_VIN_P_FC) /* PF uses 3 bits */
533 #define FILTER_SEL_WIDTH_VLD_TAG_P_FC (1 + FILTER_SEL_WIDTH_TAG_P_FC)
535 static unsigned int select_ntuple(struct c4iw_dev *dev, struct dst_entry *dst,
536 struct l2t_entry *l2t)
538 unsigned int ntuple = 0;
541 switch (dev->rdev.lldi.filt_mode) {
543 /* default filter mode */
544 case HW_TPL_FR_MT_PR_IV_P_FC:
545 if (l2t->vlan == VLAN_NONE)
546 ntuple |= FILTER_SEL_VLAN_NONE << FILTER_SEL_WIDTH_P_FC;
548 ntuple |= l2t->vlan << FILTER_SEL_WIDTH_P_FC;
549 ntuple |= 1 << FILTER_SEL_WIDTH_TAG_P_FC;
551 ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
552 FILTER_SEL_WIDTH_VLD_TAG_P_FC;
554 case HW_TPL_FR_MT_PR_OV_P_FC: {
555 viid = cxgb4_port_viid(l2t->neigh->dev);
557 ntuple |= FW_VIID_VIN_GET(viid) << FILTER_SEL_WIDTH_P_FC;
558 ntuple |= FW_VIID_PFN_GET(viid) << FILTER_SEL_WIDTH_VIN_P_FC;
559 ntuple |= FW_VIID_VIVLD_GET(viid) << FILTER_SEL_WIDTH_TAG_P_FC;
560 ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
561 FILTER_SEL_WIDTH_VLD_TAG_P_FC;
570 static int send_connect(struct c4iw_ep *ep)
572 struct cpl_act_open_req *req;
573 struct cpl_t5_act_open_req *t5_req;
574 struct cpl_act_open_req6 *req6;
575 struct cpl_t5_act_open_req6 *t5_req6;
579 unsigned int mtu_idx;
582 int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
583 sizeof(struct cpl_act_open_req) :
584 sizeof(struct cpl_t5_act_open_req);
585 int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
586 sizeof(struct cpl_act_open_req6) :
587 sizeof(struct cpl_t5_act_open_req6);
588 struct sockaddr_in *la = (struct sockaddr_in *)&ep->com.local_addr;
589 struct sockaddr_in *ra = (struct sockaddr_in *)&ep->com.remote_addr;
590 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
591 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
593 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
594 roundup(sizev4, 16) :
597 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
599 skb = get_skb(NULL, wrlen, GFP_KERNEL);
601 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
605 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
607 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
608 wscale = compute_wscale(rcv_win);
609 opt0 = (nocong ? NO_CONG(1) : 0) |
614 L2T_IDX(ep->l2t->idx) |
615 TX_CHAN(ep->tx_chan) |
616 SMAC_SEL(ep->smac_idx) |
618 ULP_MODE(ULP_MODE_TCPDDP) |
619 RCV_BUFSIZ(rcv_win>>10);
620 opt2 = RX_CHANNEL(0) |
621 CCTRL_ECN(enable_ecn) |
622 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
623 if (enable_tcp_timestamps)
624 opt2 |= TSTAMPS_EN(1);
627 if (wscale && enable_tcp_window_scaling)
628 opt2 |= WND_SCALE_EN(1);
629 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
631 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
632 if (ep->com.remote_addr.ss_family == AF_INET) {
633 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
635 OPCODE_TID(req) = cpu_to_be32(
636 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
637 ((ep->rss_qid << 14) | ep->atid)));
638 req->local_port = la->sin_port;
639 req->peer_port = ra->sin_port;
640 req->local_ip = la->sin_addr.s_addr;
641 req->peer_ip = ra->sin_addr.s_addr;
642 req->opt0 = cpu_to_be64(opt0);
643 req->params = cpu_to_be32(select_ntuple(ep->com.dev,
645 req->opt2 = cpu_to_be32(opt2);
647 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
650 OPCODE_TID(req6) = cpu_to_be32(
651 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
652 ((ep->rss_qid<<14)|ep->atid)));
653 req6->local_port = la6->sin6_port;
654 req6->peer_port = ra6->sin6_port;
655 req6->local_ip_hi = *((__be64 *)
656 (la6->sin6_addr.s6_addr));
657 req6->local_ip_lo = *((__be64 *)
658 (la6->sin6_addr.s6_addr + 8));
659 req6->peer_ip_hi = *((__be64 *)
660 (ra6->sin6_addr.s6_addr));
661 req6->peer_ip_lo = *((__be64 *)
662 (ra6->sin6_addr.s6_addr + 8));
663 req6->opt0 = cpu_to_be64(opt0);
664 req6->params = cpu_to_be32(
665 select_ntuple(ep->com.dev, ep->dst,
667 req6->opt2 = cpu_to_be32(opt2);
670 if (ep->com.remote_addr.ss_family == AF_INET) {
671 t5_req = (struct cpl_t5_act_open_req *)
673 INIT_TP_WR(t5_req, 0);
674 OPCODE_TID(t5_req) = cpu_to_be32(
675 MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
676 ((ep->rss_qid << 14) | ep->atid)));
677 t5_req->local_port = la->sin_port;
678 t5_req->peer_port = ra->sin_port;
679 t5_req->local_ip = la->sin_addr.s_addr;
680 t5_req->peer_ip = ra->sin_addr.s_addr;
681 t5_req->opt0 = cpu_to_be64(opt0);
682 t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
683 select_ntuple(ep->com.dev,
685 t5_req->opt2 = cpu_to_be32(opt2);
687 t5_req6 = (struct cpl_t5_act_open_req6 *)
689 INIT_TP_WR(t5_req6, 0);
690 OPCODE_TID(t5_req6) = cpu_to_be32(
691 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
692 ((ep->rss_qid<<14)|ep->atid)));
693 t5_req6->local_port = la6->sin6_port;
694 t5_req6->peer_port = ra6->sin6_port;
695 t5_req6->local_ip_hi = *((__be64 *)
696 (la6->sin6_addr.s6_addr));
697 t5_req6->local_ip_lo = *((__be64 *)
698 (la6->sin6_addr.s6_addr + 8));
699 t5_req6->peer_ip_hi = *((__be64 *)
700 (ra6->sin6_addr.s6_addr));
701 t5_req6->peer_ip_lo = *((__be64 *)
702 (ra6->sin6_addr.s6_addr + 8));
703 t5_req6->opt0 = cpu_to_be64(opt0);
704 t5_req6->params = (__force __be64)cpu_to_be32(
705 select_ntuple(ep->com.dev, ep->dst, ep->l2t));
706 t5_req6->opt2 = cpu_to_be32(opt2);
710 set_bit(ACT_OPEN_REQ, &ep->com.history);
711 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
714 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
718 struct fw_ofld_tx_data_wr *req;
719 struct mpa_message *mpa;
720 struct mpa_v2_conn_params mpa_v2_params;
722 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
724 BUG_ON(skb_cloned(skb));
726 mpalen = sizeof(*mpa) + ep->plen;
727 if (mpa_rev_to_use == 2)
728 mpalen += sizeof(struct mpa_v2_conn_params);
729 wrlen = roundup(mpalen + sizeof *req, 16);
730 skb = get_skb(skb, wrlen, GFP_KERNEL);
732 connect_reply_upcall(ep, -ENOMEM);
735 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
737 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
738 memset(req, 0, wrlen);
739 req->op_to_immdlen = cpu_to_be32(
740 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
742 FW_WR_IMMDLEN(mpalen));
743 req->flowid_len16 = cpu_to_be32(
744 FW_WR_FLOWID(ep->hwtid) |
745 FW_WR_LEN16(wrlen >> 4));
746 req->plen = cpu_to_be32(mpalen);
747 req->tunnel_to_proxy = cpu_to_be32(
748 FW_OFLD_TX_DATA_WR_FLUSH(1) |
749 FW_OFLD_TX_DATA_WR_SHOVE(1));
751 mpa = (struct mpa_message *)(req + 1);
752 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
753 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
754 (markers_enabled ? MPA_MARKERS : 0) |
755 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
756 mpa->private_data_size = htons(ep->plen);
757 mpa->revision = mpa_rev_to_use;
758 if (mpa_rev_to_use == 1) {
759 ep->tried_with_mpa_v1 = 1;
760 ep->retry_with_mpa_v1 = 0;
763 if (mpa_rev_to_use == 2) {
764 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
765 sizeof (struct mpa_v2_conn_params));
766 mpa_v2_params.ird = htons((u16)ep->ird);
767 mpa_v2_params.ord = htons((u16)ep->ord);
770 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
771 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
773 htons(MPA_V2_RDMA_WRITE_RTR);
774 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
776 htons(MPA_V2_RDMA_READ_RTR);
778 memcpy(mpa->private_data, &mpa_v2_params,
779 sizeof(struct mpa_v2_conn_params));
782 memcpy(mpa->private_data +
783 sizeof(struct mpa_v2_conn_params),
784 ep->mpa_pkt + sizeof(*mpa), ep->plen);
787 memcpy(mpa->private_data,
788 ep->mpa_pkt + sizeof(*mpa), ep->plen);
791 * Reference the mpa skb. This ensures the data area
792 * will remain in memory until the hw acks the tx.
793 * Function fw4_ack() will deref it.
796 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
799 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
801 state_set(&ep->com, MPA_REQ_SENT);
802 ep->mpa_attr.initiator = 1;
806 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
809 struct fw_ofld_tx_data_wr *req;
810 struct mpa_message *mpa;
812 struct mpa_v2_conn_params mpa_v2_params;
814 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
816 mpalen = sizeof(*mpa) + plen;
817 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
818 mpalen += sizeof(struct mpa_v2_conn_params);
819 wrlen = roundup(mpalen + sizeof *req, 16);
821 skb = get_skb(NULL, wrlen, GFP_KERNEL);
823 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
826 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
828 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
829 memset(req, 0, wrlen);
830 req->op_to_immdlen = cpu_to_be32(
831 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
833 FW_WR_IMMDLEN(mpalen));
834 req->flowid_len16 = cpu_to_be32(
835 FW_WR_FLOWID(ep->hwtid) |
836 FW_WR_LEN16(wrlen >> 4));
837 req->plen = cpu_to_be32(mpalen);
838 req->tunnel_to_proxy = cpu_to_be32(
839 FW_OFLD_TX_DATA_WR_FLUSH(1) |
840 FW_OFLD_TX_DATA_WR_SHOVE(1));
842 mpa = (struct mpa_message *)(req + 1);
843 memset(mpa, 0, sizeof(*mpa));
844 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
845 mpa->flags = MPA_REJECT;
846 mpa->revision = ep->mpa_attr.version;
847 mpa->private_data_size = htons(plen);
849 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
850 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
851 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
852 sizeof (struct mpa_v2_conn_params));
853 mpa_v2_params.ird = htons(((u16)ep->ird) |
854 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
856 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
858 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
859 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
860 FW_RI_INIT_P2PTYPE_READ_REQ ?
861 MPA_V2_RDMA_READ_RTR : 0) : 0));
862 memcpy(mpa->private_data, &mpa_v2_params,
863 sizeof(struct mpa_v2_conn_params));
866 memcpy(mpa->private_data +
867 sizeof(struct mpa_v2_conn_params), pdata, plen);
870 memcpy(mpa->private_data, pdata, plen);
873 * Reference the mpa skb again. This ensures the data area
874 * will remain in memory until the hw acks the tx.
875 * Function fw4_ack() will deref it.
878 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
879 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
882 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
885 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
888 struct fw_ofld_tx_data_wr *req;
889 struct mpa_message *mpa;
891 struct mpa_v2_conn_params mpa_v2_params;
893 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
895 mpalen = sizeof(*mpa) + plen;
896 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
897 mpalen += sizeof(struct mpa_v2_conn_params);
898 wrlen = roundup(mpalen + sizeof *req, 16);
900 skb = get_skb(NULL, wrlen, GFP_KERNEL);
902 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
905 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
907 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
908 memset(req, 0, wrlen);
909 req->op_to_immdlen = cpu_to_be32(
910 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
912 FW_WR_IMMDLEN(mpalen));
913 req->flowid_len16 = cpu_to_be32(
914 FW_WR_FLOWID(ep->hwtid) |
915 FW_WR_LEN16(wrlen >> 4));
916 req->plen = cpu_to_be32(mpalen);
917 req->tunnel_to_proxy = cpu_to_be32(
918 FW_OFLD_TX_DATA_WR_FLUSH(1) |
919 FW_OFLD_TX_DATA_WR_SHOVE(1));
921 mpa = (struct mpa_message *)(req + 1);
922 memset(mpa, 0, sizeof(*mpa));
923 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
924 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
925 (markers_enabled ? MPA_MARKERS : 0);
926 mpa->revision = ep->mpa_attr.version;
927 mpa->private_data_size = htons(plen);
929 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
930 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
931 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
932 sizeof (struct mpa_v2_conn_params));
933 mpa_v2_params.ird = htons((u16)ep->ird);
934 mpa_v2_params.ord = htons((u16)ep->ord);
935 if (peer2peer && (ep->mpa_attr.p2p_type !=
936 FW_RI_INIT_P2PTYPE_DISABLED)) {
937 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
939 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
941 htons(MPA_V2_RDMA_WRITE_RTR);
942 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
944 htons(MPA_V2_RDMA_READ_RTR);
947 memcpy(mpa->private_data, &mpa_v2_params,
948 sizeof(struct mpa_v2_conn_params));
951 memcpy(mpa->private_data +
952 sizeof(struct mpa_v2_conn_params), pdata, plen);
955 memcpy(mpa->private_data, pdata, plen);
958 * Reference the mpa skb. This ensures the data area
959 * will remain in memory until the hw acks the tx.
960 * Function fw4_ack() will deref it.
963 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
965 state_set(&ep->com, MPA_REP_SENT);
966 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
969 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
972 struct cpl_act_establish *req = cplhdr(skb);
973 unsigned int tid = GET_TID(req);
974 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
975 struct tid_info *t = dev->rdev.lldi.tids;
977 ep = lookup_atid(t, atid);
979 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
980 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
982 dst_confirm(ep->dst);
984 /* setup the hwtid for this connection */
986 cxgb4_insert_tid(t, ep, tid);
987 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
989 ep->snd_seq = be32_to_cpu(req->snd_isn);
990 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
992 set_emss(ep, ntohs(req->tcp_opt));
994 /* dealloc the atid */
995 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
996 cxgb4_free_atid(t, atid);
997 set_bit(ACT_ESTAB, &ep->com.history);
999 /* start MPA negotiation */
1000 send_flowc(ep, NULL);
1001 if (ep->retry_with_mpa_v1)
1002 send_mpa_req(ep, skb, 1);
1004 send_mpa_req(ep, skb, mpa_rev);
1009 static void close_complete_upcall(struct c4iw_ep *ep)
1011 struct iw_cm_event event;
1013 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1014 memset(&event, 0, sizeof(event));
1015 event.event = IW_CM_EVENT_CLOSE;
1016 if (ep->com.cm_id) {
1017 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1018 ep, ep->com.cm_id, ep->hwtid);
1019 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1020 ep->com.cm_id->rem_ref(ep->com.cm_id);
1021 ep->com.cm_id = NULL;
1022 set_bit(CLOSE_UPCALL, &ep->com.history);
1026 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
1028 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1029 close_complete_upcall(ep);
1030 state_set(&ep->com, ABORTING);
1031 set_bit(ABORT_CONN, &ep->com.history);
1032 return send_abort(ep, skb, gfp);
1035 static void peer_close_upcall(struct c4iw_ep *ep)
1037 struct iw_cm_event event;
1039 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1040 memset(&event, 0, sizeof(event));
1041 event.event = IW_CM_EVENT_DISCONNECT;
1042 if (ep->com.cm_id) {
1043 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1044 ep, ep->com.cm_id, ep->hwtid);
1045 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1046 set_bit(DISCONN_UPCALL, &ep->com.history);
1050 static void peer_abort_upcall(struct c4iw_ep *ep)
1052 struct iw_cm_event event;
1054 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1055 memset(&event, 0, sizeof(event));
1056 event.event = IW_CM_EVENT_CLOSE;
1057 event.status = -ECONNRESET;
1058 if (ep->com.cm_id) {
1059 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1060 ep->com.cm_id, ep->hwtid);
1061 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1062 ep->com.cm_id->rem_ref(ep->com.cm_id);
1063 ep->com.cm_id = NULL;
1064 set_bit(ABORT_UPCALL, &ep->com.history);
1068 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1070 struct iw_cm_event event;
1072 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1073 memset(&event, 0, sizeof(event));
1074 event.event = IW_CM_EVENT_CONNECT_REPLY;
1075 event.status = status;
1076 memcpy(&event.local_addr, &ep->com.local_addr,
1077 sizeof(ep->com.local_addr));
1078 memcpy(&event.remote_addr, &ep->com.remote_addr,
1079 sizeof(ep->com.remote_addr));
1081 if ((status == 0) || (status == -ECONNREFUSED)) {
1082 if (!ep->tried_with_mpa_v1) {
1083 /* this means MPA_v2 is used */
1084 event.private_data_len = ep->plen -
1085 sizeof(struct mpa_v2_conn_params);
1086 event.private_data = ep->mpa_pkt +
1087 sizeof(struct mpa_message) +
1088 sizeof(struct mpa_v2_conn_params);
1090 /* this means MPA_v1 is used */
1091 event.private_data_len = ep->plen;
1092 event.private_data = ep->mpa_pkt +
1093 sizeof(struct mpa_message);
1097 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1099 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1100 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1103 ep->com.cm_id->rem_ref(ep->com.cm_id);
1104 ep->com.cm_id = NULL;
1108 static void connect_request_upcall(struct c4iw_ep *ep)
1110 struct iw_cm_event event;
1112 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1113 memset(&event, 0, sizeof(event));
1114 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1115 memcpy(&event.local_addr, &ep->com.local_addr,
1116 sizeof(ep->com.local_addr));
1117 memcpy(&event.remote_addr, &ep->com.remote_addr,
1118 sizeof(ep->com.remote_addr));
1119 event.provider_data = ep;
1120 if (!ep->tried_with_mpa_v1) {
1121 /* this means MPA_v2 is used */
1122 event.ord = ep->ord;
1123 event.ird = ep->ird;
1124 event.private_data_len = ep->plen -
1125 sizeof(struct mpa_v2_conn_params);
1126 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1127 sizeof(struct mpa_v2_conn_params);
1129 /* this means MPA_v1 is used. Send max supported */
1130 event.ord = c4iw_max_read_depth;
1131 event.ird = c4iw_max_read_depth;
1132 event.private_data_len = ep->plen;
1133 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1135 if (state_read(&ep->parent_ep->com) != DEAD) {
1136 c4iw_get_ep(&ep->com);
1137 ep->parent_ep->com.cm_id->event_handler(
1138 ep->parent_ep->com.cm_id,
1141 set_bit(CONNREQ_UPCALL, &ep->com.history);
1142 c4iw_put_ep(&ep->parent_ep->com);
1143 ep->parent_ep = NULL;
1146 static void established_upcall(struct c4iw_ep *ep)
1148 struct iw_cm_event event;
1150 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1151 memset(&event, 0, sizeof(event));
1152 event.event = IW_CM_EVENT_ESTABLISHED;
1153 event.ird = ep->ird;
1154 event.ord = ep->ord;
1155 if (ep->com.cm_id) {
1156 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1157 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1158 set_bit(ESTAB_UPCALL, &ep->com.history);
1162 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1164 struct cpl_rx_data_ack *req;
1165 struct sk_buff *skb;
1166 int wrlen = roundup(sizeof *req, 16);
1168 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1169 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1171 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1175 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1176 memset(req, 0, wrlen);
1177 INIT_TP_WR(req, ep->hwtid);
1178 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1180 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1182 V_RX_DACK_MODE(dack_mode));
1183 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1184 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1188 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1190 struct mpa_message *mpa;
1191 struct mpa_v2_conn_params *mpa_v2_params;
1193 u16 resp_ird, resp_ord;
1194 u8 rtr_mismatch = 0, insuff_ird = 0;
1195 struct c4iw_qp_attributes attrs;
1196 enum c4iw_qp_attr_mask mask;
1199 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1202 * Stop mpa timer. If it expired, then the state has
1203 * changed and we bail since ep_timeout already aborted
1207 if (state_read(&ep->com) != MPA_REQ_SENT)
1211 * If we get more than the supported amount of private data
1212 * then we must fail this connection.
1214 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1220 * copy the new data into our accumulation buffer.
1222 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1224 ep->mpa_pkt_len += skb->len;
1227 * if we don't even have the mpa message, then bail.
1229 if (ep->mpa_pkt_len < sizeof(*mpa))
1231 mpa = (struct mpa_message *) ep->mpa_pkt;
1233 /* Validate MPA header. */
1234 if (mpa->revision > mpa_rev) {
1235 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1236 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1240 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1245 plen = ntohs(mpa->private_data_size);
1248 * Fail if there's too much private data.
1250 if (plen > MPA_MAX_PRIVATE_DATA) {
1256 * If plen does not account for pkt size
1258 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1263 ep->plen = (u8) plen;
1266 * If we don't have all the pdata yet, then bail.
1267 * We'll continue process when more data arrives.
1269 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1272 if (mpa->flags & MPA_REJECT) {
1273 err = -ECONNREFUSED;
1278 * If we get here we have accumulated the entire mpa
1279 * start reply message including private data. And
1280 * the MPA header is valid.
1282 state_set(&ep->com, FPDU_MODE);
1283 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1284 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1285 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1286 ep->mpa_attr.version = mpa->revision;
1287 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1289 if (mpa->revision == 2) {
1290 ep->mpa_attr.enhanced_rdma_conn =
1291 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1292 if (ep->mpa_attr.enhanced_rdma_conn) {
1293 mpa_v2_params = (struct mpa_v2_conn_params *)
1294 (ep->mpa_pkt + sizeof(*mpa));
1295 resp_ird = ntohs(mpa_v2_params->ird) &
1296 MPA_V2_IRD_ORD_MASK;
1297 resp_ord = ntohs(mpa_v2_params->ord) &
1298 MPA_V2_IRD_ORD_MASK;
1301 * This is a double-check. Ideally, below checks are
1302 * not required since ird/ord stuff has been taken
1303 * care of in c4iw_accept_cr
1305 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1312 if (ntohs(mpa_v2_params->ird) &
1313 MPA_V2_PEER2PEER_MODEL) {
1314 if (ntohs(mpa_v2_params->ord) &
1315 MPA_V2_RDMA_WRITE_RTR)
1316 ep->mpa_attr.p2p_type =
1317 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1318 else if (ntohs(mpa_v2_params->ord) &
1319 MPA_V2_RDMA_READ_RTR)
1320 ep->mpa_attr.p2p_type =
1321 FW_RI_INIT_P2PTYPE_READ_REQ;
1324 } else if (mpa->revision == 1)
1326 ep->mpa_attr.p2p_type = p2p_type;
1328 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1329 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1330 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1331 ep->mpa_attr.recv_marker_enabled,
1332 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1333 ep->mpa_attr.p2p_type, p2p_type);
1336 * If responder's RTR does not match with that of initiator, assign
1337 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1338 * generated when moving QP to RTS state.
1339 * A TERM message will be sent after QP has moved to RTS state
1341 if ((ep->mpa_attr.version == 2) && peer2peer &&
1342 (ep->mpa_attr.p2p_type != p2p_type)) {
1343 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1347 attrs.mpa_attr = ep->mpa_attr;
1348 attrs.max_ird = ep->ird;
1349 attrs.max_ord = ep->ord;
1350 attrs.llp_stream_handle = ep;
1351 attrs.next_state = C4IW_QP_STATE_RTS;
1353 mask = C4IW_QP_ATTR_NEXT_STATE |
1354 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1355 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1357 /* bind QP and TID with INIT_WR */
1358 err = c4iw_modify_qp(ep->com.qp->rhp,
1359 ep->com.qp, mask, &attrs, 1);
1364 * If responder's RTR requirement did not match with what initiator
1365 * supports, generate TERM message
1368 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1369 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1370 attrs.ecode = MPA_NOMATCH_RTR;
1371 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1372 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1373 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1379 * Generate TERM if initiator IRD is not sufficient for responder
1380 * provided ORD. Currently, we do the same behaviour even when
1381 * responder provided IRD is also not sufficient as regards to
1385 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1387 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1388 attrs.ecode = MPA_INSUFF_IRD;
1389 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1390 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1391 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1397 state_set(&ep->com, ABORTING);
1398 send_abort(ep, skb, GFP_KERNEL);
1400 connect_reply_upcall(ep, err);
1404 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1406 struct mpa_message *mpa;
1407 struct mpa_v2_conn_params *mpa_v2_params;
1410 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1412 if (state_read(&ep->com) != MPA_REQ_WAIT)
1416 * If we get more than the supported amount of private data
1417 * then we must fail this connection.
1419 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1421 abort_connection(ep, skb, GFP_KERNEL);
1425 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1428 * Copy the new data into our accumulation buffer.
1430 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1432 ep->mpa_pkt_len += skb->len;
1435 * If we don't even have the mpa message, then bail.
1436 * We'll continue process when more data arrives.
1438 if (ep->mpa_pkt_len < sizeof(*mpa))
1441 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1443 mpa = (struct mpa_message *) ep->mpa_pkt;
1446 * Validate MPA Header.
1448 if (mpa->revision > mpa_rev) {
1449 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1450 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1452 abort_connection(ep, skb, GFP_KERNEL);
1456 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1458 abort_connection(ep, skb, GFP_KERNEL);
1462 plen = ntohs(mpa->private_data_size);
1465 * Fail if there's too much private data.
1467 if (plen > MPA_MAX_PRIVATE_DATA) {
1469 abort_connection(ep, skb, GFP_KERNEL);
1474 * If plen does not account for pkt size
1476 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1478 abort_connection(ep, skb, GFP_KERNEL);
1481 ep->plen = (u8) plen;
1484 * If we don't have all the pdata yet, then bail.
1486 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1490 * If we get here we have accumulated the entire mpa
1491 * start reply message including private data.
1493 ep->mpa_attr.initiator = 0;
1494 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1495 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1496 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1497 ep->mpa_attr.version = mpa->revision;
1498 if (mpa->revision == 1)
1499 ep->tried_with_mpa_v1 = 1;
1500 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1502 if (mpa->revision == 2) {
1503 ep->mpa_attr.enhanced_rdma_conn =
1504 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1505 if (ep->mpa_attr.enhanced_rdma_conn) {
1506 mpa_v2_params = (struct mpa_v2_conn_params *)
1507 (ep->mpa_pkt + sizeof(*mpa));
1508 ep->ird = ntohs(mpa_v2_params->ird) &
1509 MPA_V2_IRD_ORD_MASK;
1510 ep->ord = ntohs(mpa_v2_params->ord) &
1511 MPA_V2_IRD_ORD_MASK;
1512 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1514 if (ntohs(mpa_v2_params->ord) &
1515 MPA_V2_RDMA_WRITE_RTR)
1516 ep->mpa_attr.p2p_type =
1517 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1518 else if (ntohs(mpa_v2_params->ord) &
1519 MPA_V2_RDMA_READ_RTR)
1520 ep->mpa_attr.p2p_type =
1521 FW_RI_INIT_P2PTYPE_READ_REQ;
1524 } else if (mpa->revision == 1)
1526 ep->mpa_attr.p2p_type = p2p_type;
1528 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1529 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1530 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1531 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1532 ep->mpa_attr.p2p_type);
1534 state_set(&ep->com, MPA_REQ_RCVD);
1537 connect_request_upcall(ep);
1541 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1544 struct cpl_rx_data *hdr = cplhdr(skb);
1545 unsigned int dlen = ntohs(hdr->len);
1546 unsigned int tid = GET_TID(hdr);
1547 struct tid_info *t = dev->rdev.lldi.tids;
1548 __u8 status = hdr->status;
1550 ep = lookup_tid(t, tid);
1551 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1552 skb_pull(skb, sizeof(*hdr));
1553 skb_trim(skb, dlen);
1555 /* update RX credits */
1556 update_rx_credits(ep, dlen);
1558 switch (state_read(&ep->com)) {
1560 ep->rcv_seq += dlen;
1561 process_mpa_reply(ep, skb);
1564 ep->rcv_seq += dlen;
1565 process_mpa_request(ep, skb);
1568 struct c4iw_qp_attributes attrs;
1569 BUG_ON(!ep->com.qp);
1571 pr_err("%s Unexpected streaming data." \
1572 " qpid %u ep %p state %d tid %u status %d\n",
1573 __func__, ep->com.qp->wq.sq.qid, ep,
1574 state_read(&ep->com), ep->hwtid, status);
1575 attrs.next_state = C4IW_QP_STATE_ERROR;
1576 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1577 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1578 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
1587 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1590 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1592 unsigned int tid = GET_TID(rpl);
1593 struct tid_info *t = dev->rdev.lldi.tids;
1595 ep = lookup_tid(t, tid);
1597 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1600 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1601 mutex_lock(&ep->com.mutex);
1602 switch (ep->com.state) {
1604 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1605 __state_set(&ep->com, DEAD);
1609 printk(KERN_ERR "%s ep %p state %d\n",
1610 __func__, ep, ep->com.state);
1613 mutex_unlock(&ep->com.mutex);
1616 release_ep_resources(ep);
1620 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1622 struct sk_buff *skb;
1623 struct fw_ofld_connection_wr *req;
1624 unsigned int mtu_idx;
1626 struct sockaddr_in *sin;
1628 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1629 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1630 memset(req, 0, sizeof(*req));
1631 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1632 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1633 req->le.filter = cpu_to_be32(select_ntuple(ep->com.dev, ep->dst,
1635 sin = (struct sockaddr_in *)&ep->com.local_addr;
1636 req->le.lport = sin->sin_port;
1637 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1638 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1639 req->le.pport = sin->sin_port;
1640 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1641 req->tcb.t_state_to_astid =
1642 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1643 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1644 req->tcb.cplrxdataack_cplpassacceptrpl =
1645 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1646 req->tcb.tx_max = (__force __be32) jiffies;
1647 req->tcb.rcv_adv = htons(1);
1648 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1649 wscale = compute_wscale(rcv_win);
1650 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1651 (nocong ? NO_CONG(1) : 0) |
1656 L2T_IDX(ep->l2t->idx) |
1657 TX_CHAN(ep->tx_chan) |
1658 SMAC_SEL(ep->smac_idx) |
1660 ULP_MODE(ULP_MODE_TCPDDP) |
1661 RCV_BUFSIZ(rcv_win >> 10));
1662 req->tcb.opt2 = (__force __be32) (PACE(1) |
1663 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1665 CCTRL_ECN(enable_ecn) |
1666 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1667 if (enable_tcp_timestamps)
1668 req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1669 if (enable_tcp_sack)
1670 req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1671 if (wscale && enable_tcp_window_scaling)
1672 req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1673 req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1674 req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1675 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1676 set_bit(ACT_OFLD_CONN, &ep->com.history);
1677 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1681 * Return whether a failed active open has allocated a TID
1683 static inline int act_open_has_tid(int status)
1685 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1686 status != CPL_ERR_ARP_MISS;
1689 #define ACT_OPEN_RETRY_COUNT 2
1691 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1692 struct dst_entry *dst, struct c4iw_dev *cdev,
1695 struct neighbour *n;
1697 struct net_device *pdev;
1699 n = dst_neigh_lookup(dst, peer_ip);
1705 if (n->dev->flags & IFF_LOOPBACK) {
1707 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1708 else if (IS_ENABLED(CONFIG_IPV6))
1709 for_each_netdev(&init_net, pdev) {
1710 if (ipv6_chk_addr(&init_net,
1711 (struct in6_addr *)peer_ip,
1722 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1726 ep->mtu = pdev->mtu;
1727 ep->tx_chan = cxgb4_port_chan(pdev);
1728 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1729 step = cdev->rdev.lldi.ntxq /
1730 cdev->rdev.lldi.nchan;
1731 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1732 step = cdev->rdev.lldi.nrxq /
1733 cdev->rdev.lldi.nchan;
1734 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1735 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1736 cxgb4_port_idx(pdev) * step];
1739 pdev = get_real_dev(n->dev);
1740 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1744 ep->mtu = dst_mtu(dst);
1745 ep->tx_chan = cxgb4_port_chan(n->dev);
1746 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1747 step = cdev->rdev.lldi.ntxq /
1748 cdev->rdev.lldi.nchan;
1749 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1750 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1751 step = cdev->rdev.lldi.nrxq /
1752 cdev->rdev.lldi.nchan;
1753 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1754 cxgb4_port_idx(n->dev) * step];
1757 ep->retry_with_mpa_v1 = 0;
1758 ep->tried_with_mpa_v1 = 0;
1770 static int c4iw_reconnect(struct c4iw_ep *ep)
1773 struct sockaddr_in *laddr = (struct sockaddr_in *)
1774 &ep->com.cm_id->local_addr;
1775 struct sockaddr_in *raddr = (struct sockaddr_in *)
1776 &ep->com.cm_id->remote_addr;
1777 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
1778 &ep->com.cm_id->local_addr;
1779 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
1780 &ep->com.cm_id->remote_addr;
1784 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1785 init_timer(&ep->timer);
1788 * Allocate an active TID to initiate a TCP connection.
1790 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1791 if (ep->atid == -1) {
1792 pr_err("%s - cannot alloc atid.\n", __func__);
1796 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
1799 if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
1800 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
1801 raddr->sin_addr.s_addr, laddr->sin_port,
1802 raddr->sin_port, 0);
1804 ra = (__u8 *)&raddr->sin_addr;
1806 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
1807 raddr6->sin6_addr.s6_addr,
1808 laddr6->sin6_port, raddr6->sin6_port, 0,
1809 raddr6->sin6_scope_id);
1811 ra = (__u8 *)&raddr6->sin6_addr;
1814 pr_err("%s - cannot find route.\n", __func__);
1815 err = -EHOSTUNREACH;
1818 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
1820 pr_err("%s - cannot alloc l2e.\n", __func__);
1824 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1825 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1828 state_set(&ep->com, CONNECTING);
1831 /* send connect request to rnic */
1832 err = send_connect(ep);
1836 cxgb4_l2t_release(ep->l2t);
1838 dst_release(ep->dst);
1840 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
1841 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1844 * remember to send notification to upper layer.
1845 * We are in here so the upper layer is not aware that this is
1846 * re-connect attempt and so, upper layer is still waiting for
1847 * response of 1st connect request.
1849 connect_reply_upcall(ep, -ECONNRESET);
1850 c4iw_put_ep(&ep->com);
1855 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1858 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1859 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1860 ntohl(rpl->atid_status)));
1861 struct tid_info *t = dev->rdev.lldi.tids;
1862 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1863 struct sockaddr_in *la;
1864 struct sockaddr_in *ra;
1865 struct sockaddr_in6 *la6;
1866 struct sockaddr_in6 *ra6;
1868 ep = lookup_atid(t, atid);
1869 la = (struct sockaddr_in *)&ep->com.local_addr;
1870 ra = (struct sockaddr_in *)&ep->com.remote_addr;
1871 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
1872 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
1874 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1875 status, status2errno(status));
1877 if (status == CPL_ERR_RTX_NEG_ADVICE) {
1878 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1883 set_bit(ACT_OPEN_RPL, &ep->com.history);
1886 * Log interesting failures.
1889 case CPL_ERR_CONN_RESET:
1890 case CPL_ERR_CONN_TIMEDOUT:
1892 case CPL_ERR_TCAM_FULL:
1893 mutex_lock(&dev->rdev.stats.lock);
1894 dev->rdev.stats.tcam_full++;
1895 mutex_unlock(&dev->rdev.stats.lock);
1896 if (ep->com.local_addr.ss_family == AF_INET &&
1897 dev->rdev.lldi.enable_fw_ofld_conn) {
1898 send_fw_act_open_req(ep,
1899 GET_TID_TID(GET_AOPEN_ATID(
1900 ntohl(rpl->atid_status))));
1904 case CPL_ERR_CONN_EXIST:
1905 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
1906 set_bit(ACT_RETRY_INUSE, &ep->com.history);
1907 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
1909 cxgb4_free_atid(t, atid);
1910 dst_release(ep->dst);
1911 cxgb4_l2t_release(ep->l2t);
1917 if (ep->com.local_addr.ss_family == AF_INET) {
1918 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1919 atid, status, status2errno(status),
1920 &la->sin_addr.s_addr, ntohs(la->sin_port),
1921 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
1923 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
1924 atid, status, status2errno(status),
1925 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
1926 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
1931 connect_reply_upcall(ep, status2errno(status));
1932 state_set(&ep->com, DEAD);
1934 if (status && act_open_has_tid(status))
1935 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1937 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1938 cxgb4_free_atid(t, atid);
1939 dst_release(ep->dst);
1940 cxgb4_l2t_release(ep->l2t);
1941 c4iw_put_ep(&ep->com);
1946 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1948 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1949 struct tid_info *t = dev->rdev.lldi.tids;
1950 unsigned int stid = GET_TID(rpl);
1951 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1954 PDBG("%s stid %d lookup failure!\n", __func__, stid);
1957 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1958 rpl->status, status2errno(rpl->status));
1959 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1965 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1967 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1968 struct tid_info *t = dev->rdev.lldi.tids;
1969 unsigned int stid = GET_TID(rpl);
1970 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1972 PDBG("%s ep %p\n", __func__, ep);
1973 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1977 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
1978 struct cpl_pass_accept_req *req)
1980 struct cpl_pass_accept_rpl *rpl;
1981 unsigned int mtu_idx;
1986 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1987 BUG_ON(skb_cloned(skb));
1988 skb_trim(skb, sizeof(*rpl));
1990 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1991 wscale = compute_wscale(rcv_win);
1992 opt0 = (nocong ? NO_CONG(1) : 0) |
1997 L2T_IDX(ep->l2t->idx) |
1998 TX_CHAN(ep->tx_chan) |
1999 SMAC_SEL(ep->smac_idx) |
2000 DSCP(ep->tos >> 2) |
2001 ULP_MODE(ULP_MODE_TCPDDP) |
2002 RCV_BUFSIZ(rcv_win>>10);
2003 opt2 = RX_CHANNEL(0) |
2004 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
2006 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2007 opt2 |= TSTAMPS_EN(1);
2008 if (enable_tcp_sack && req->tcpopt.sack)
2010 if (wscale && enable_tcp_window_scaling)
2011 opt2 |= WND_SCALE_EN(1);
2013 const struct tcphdr *tcph;
2014 u32 hlen = ntohl(req->hdr_len);
2016 tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
2018 if (tcph->ece && tcph->cwr)
2019 opt2 |= CCTRL_ECN(1);
2023 INIT_TP_WR(rpl, ep->hwtid);
2024 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2026 rpl->opt0 = cpu_to_be64(opt0);
2027 rpl->opt2 = cpu_to_be32(opt2);
2028 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2029 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2034 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2036 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2037 BUG_ON(skb_cloned(skb));
2038 skb_trim(skb, sizeof(struct cpl_tid_release));
2040 release_tid(&dev->rdev, hwtid, skb);
2044 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
2045 __u8 *local_ip, __u8 *peer_ip,
2046 __be16 *local_port, __be16 *peer_port)
2048 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
2049 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
2050 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2051 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2052 struct tcphdr *tcp = (struct tcphdr *)
2053 ((u8 *)(req + 1) + eth_len + ip_len);
2055 if (ip->version == 4) {
2056 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2057 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2060 memcpy(peer_ip, &ip->saddr, 4);
2061 memcpy(local_ip, &ip->daddr, 4);
2063 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2064 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2067 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2068 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2070 *peer_port = tcp->source;
2071 *local_port = tcp->dest;
2076 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2078 struct c4iw_ep *child_ep = NULL, *parent_ep;
2079 struct cpl_pass_accept_req *req = cplhdr(skb);
2080 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
2081 struct tid_info *t = dev->rdev.lldi.tids;
2082 unsigned int hwtid = GET_TID(req);
2083 struct dst_entry *dst;
2084 __u8 local_ip[16], peer_ip[16];
2085 __be16 local_port, peer_port;
2087 u16 peer_mss = ntohs(req->tcpopt.mss);
2090 parent_ep = lookup_stid(t, stid);
2092 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2096 if (state_read(&parent_ep->com) != LISTEN) {
2097 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
2102 get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
2104 /* Find output route */
2106 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2107 , __func__, parent_ep, hwtid,
2108 local_ip, peer_ip, ntohs(local_port),
2109 ntohs(peer_port), peer_mss);
2110 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2111 local_port, peer_port,
2112 GET_POPEN_TOS(ntohl(req->tos_stid)));
2114 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2115 , __func__, parent_ep, hwtid,
2116 local_ip, peer_ip, ntohs(local_port),
2117 ntohs(peer_port), peer_mss);
2118 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2119 PASS_OPEN_TOS(ntohl(req->tos_stid)),
2120 ((struct sockaddr_in6 *)
2121 &parent_ep->com.local_addr)->sin6_scope_id);
2124 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2129 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2131 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2137 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
2139 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2146 if (peer_mss && child_ep->mtu > (peer_mss + 40))
2147 child_ep->mtu = peer_mss + 40;
2149 state_set(&child_ep->com, CONNECTING);
2150 child_ep->com.dev = dev;
2151 child_ep->com.cm_id = NULL;
2153 struct sockaddr_in *sin = (struct sockaddr_in *)
2154 &child_ep->com.local_addr;
2155 sin->sin_family = PF_INET;
2156 sin->sin_port = local_port;
2157 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2158 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2159 sin->sin_family = PF_INET;
2160 sin->sin_port = peer_port;
2161 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2163 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
2164 &child_ep->com.local_addr;
2165 sin6->sin6_family = PF_INET6;
2166 sin6->sin6_port = local_port;
2167 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2168 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2169 sin6->sin6_family = PF_INET6;
2170 sin6->sin6_port = peer_port;
2171 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2173 c4iw_get_ep(&parent_ep->com);
2174 child_ep->parent_ep = parent_ep;
2175 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
2176 child_ep->dst = dst;
2177 child_ep->hwtid = hwtid;
2179 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2180 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2182 init_timer(&child_ep->timer);
2183 cxgb4_insert_tid(t, child_ep, hwtid);
2184 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2185 accept_cr(child_ep, skb, req);
2186 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2189 reject_cr(dev, hwtid, skb);
2194 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2197 struct cpl_pass_establish *req = cplhdr(skb);
2198 struct tid_info *t = dev->rdev.lldi.tids;
2199 unsigned int tid = GET_TID(req);
2201 ep = lookup_tid(t, tid);
2202 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2203 ep->snd_seq = be32_to_cpu(req->snd_isn);
2204 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2206 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2207 ntohs(req->tcp_opt));
2209 set_emss(ep, ntohs(req->tcp_opt));
2211 dst_confirm(ep->dst);
2212 state_set(&ep->com, MPA_REQ_WAIT);
2214 send_flowc(ep, skb);
2215 set_bit(PASS_ESTAB, &ep->com.history);
2220 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2222 struct cpl_peer_close *hdr = cplhdr(skb);
2224 struct c4iw_qp_attributes attrs;
2227 struct tid_info *t = dev->rdev.lldi.tids;
2228 unsigned int tid = GET_TID(hdr);
2231 ep = lookup_tid(t, tid);
2232 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2233 dst_confirm(ep->dst);
2235 set_bit(PEER_CLOSE, &ep->com.history);
2236 mutex_lock(&ep->com.mutex);
2237 switch (ep->com.state) {
2239 __state_set(&ep->com, CLOSING);
2242 __state_set(&ep->com, CLOSING);
2243 connect_reply_upcall(ep, -ECONNRESET);
2248 * We're gonna mark this puppy DEAD, but keep
2249 * the reference on it until the ULP accepts or
2250 * rejects the CR. Also wake up anyone waiting
2251 * in rdma connection migration (see c4iw_accept_cr()).
2253 __state_set(&ep->com, CLOSING);
2254 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2255 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2258 __state_set(&ep->com, CLOSING);
2259 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2260 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2264 __state_set(&ep->com, CLOSING);
2265 attrs.next_state = C4IW_QP_STATE_CLOSING;
2266 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2267 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2268 if (ret != -ECONNRESET) {
2269 peer_close_upcall(ep);
2277 __state_set(&ep->com, MORIBUND);
2282 if (ep->com.cm_id && ep->com.qp) {
2283 attrs.next_state = C4IW_QP_STATE_IDLE;
2284 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2285 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2287 close_complete_upcall(ep);
2288 __state_set(&ep->com, DEAD);
2298 mutex_unlock(&ep->com.mutex);
2300 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2302 release_ep_resources(ep);
2307 * Returns whether an ABORT_REQ_RSS message is a negative advice.
2309 static int is_neg_adv_abort(unsigned int status)
2311 return status == CPL_ERR_RTX_NEG_ADVICE ||
2312 status == CPL_ERR_PERSIST_NEG_ADVICE;
2315 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2317 struct cpl_abort_req_rss *req = cplhdr(skb);
2319 struct cpl_abort_rpl *rpl;
2320 struct sk_buff *rpl_skb;
2321 struct c4iw_qp_attributes attrs;
2324 struct tid_info *t = dev->rdev.lldi.tids;
2325 unsigned int tid = GET_TID(req);
2327 ep = lookup_tid(t, tid);
2328 if (is_neg_adv_abort(req->status)) {
2329 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2333 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2335 set_bit(PEER_ABORT, &ep->com.history);
2338 * Wake up any threads in rdma_init() or rdma_fini().
2339 * However, this is not needed if com state is just
2342 if (ep->com.state != MPA_REQ_SENT)
2343 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2345 mutex_lock(&ep->com.mutex);
2346 switch (ep->com.state) {
2354 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2355 connect_reply_upcall(ep, -ECONNRESET);
2358 * we just don't send notification upwards because we
2359 * want to retry with mpa_v1 without upper layers even
2362 * do some housekeeping so as to re-initiate the
2365 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2367 ep->retry_with_mpa_v1 = 1;
2379 if (ep->com.cm_id && ep->com.qp) {
2380 attrs.next_state = C4IW_QP_STATE_ERROR;
2381 ret = c4iw_modify_qp(ep->com.qp->rhp,
2382 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2386 "%s - qp <- error failed!\n",
2389 peer_abort_upcall(ep);
2394 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2395 mutex_unlock(&ep->com.mutex);
2401 dst_confirm(ep->dst);
2402 if (ep->com.state != ABORTING) {
2403 __state_set(&ep->com, DEAD);
2404 /* we don't release if we want to retry with mpa_v1 */
2405 if (!ep->retry_with_mpa_v1)
2408 mutex_unlock(&ep->com.mutex);
2410 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2412 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2417 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2418 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2419 INIT_TP_WR(rpl, ep->hwtid);
2420 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2421 rpl->cmd = CPL_ABORT_NO_RST;
2422 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2425 release_ep_resources(ep);
2426 else if (ep->retry_with_mpa_v1) {
2427 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2428 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2429 dst_release(ep->dst);
2430 cxgb4_l2t_release(ep->l2t);
2437 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2440 struct c4iw_qp_attributes attrs;
2441 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2443 struct tid_info *t = dev->rdev.lldi.tids;
2444 unsigned int tid = GET_TID(rpl);
2446 ep = lookup_tid(t, tid);
2448 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2451 /* The cm_id may be null if we failed to connect */
2452 mutex_lock(&ep->com.mutex);
2453 switch (ep->com.state) {
2455 __state_set(&ep->com, MORIBUND);
2459 if ((ep->com.cm_id) && (ep->com.qp)) {
2460 attrs.next_state = C4IW_QP_STATE_IDLE;
2461 c4iw_modify_qp(ep->com.qp->rhp,
2463 C4IW_QP_ATTR_NEXT_STATE,
2466 close_complete_upcall(ep);
2467 __state_set(&ep->com, DEAD);
2477 mutex_unlock(&ep->com.mutex);
2479 release_ep_resources(ep);
2483 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2485 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2486 struct tid_info *t = dev->rdev.lldi.tids;
2487 unsigned int tid = GET_TID(rpl);
2489 struct c4iw_qp_attributes attrs;
2491 ep = lookup_tid(t, tid);
2494 if (ep && ep->com.qp) {
2495 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2496 ep->com.qp->wq.sq.qid);
2497 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2498 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2499 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2501 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2507 * Upcall from the adapter indicating data has been transmitted.
2508 * For us its just the single MPA request or reply. We can now free
2509 * the skb holding the mpa message.
2511 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2514 struct cpl_fw4_ack *hdr = cplhdr(skb);
2515 u8 credits = hdr->credits;
2516 unsigned int tid = GET_TID(hdr);
2517 struct tid_info *t = dev->rdev.lldi.tids;
2520 ep = lookup_tid(t, tid);
2521 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2523 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2524 __func__, ep, ep->hwtid, state_read(&ep->com));
2528 dst_confirm(ep->dst);
2530 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2531 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2532 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2533 kfree_skb(ep->mpa_skb);
2539 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2542 struct c4iw_ep *ep = to_ep(cm_id);
2543 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2545 if (state_read(&ep->com) == DEAD) {
2546 c4iw_put_ep(&ep->com);
2549 set_bit(ULP_REJECT, &ep->com.history);
2550 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2552 abort_connection(ep, NULL, GFP_KERNEL);
2554 err = send_mpa_reject(ep, pdata, pdata_len);
2555 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2557 c4iw_put_ep(&ep->com);
2561 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2564 struct c4iw_qp_attributes attrs;
2565 enum c4iw_qp_attr_mask mask;
2566 struct c4iw_ep *ep = to_ep(cm_id);
2567 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2568 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2570 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2571 if (state_read(&ep->com) == DEAD) {
2576 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2579 set_bit(ULP_ACCEPT, &ep->com.history);
2580 if ((conn_param->ord > c4iw_max_read_depth) ||
2581 (conn_param->ird > c4iw_max_read_depth)) {
2582 abort_connection(ep, NULL, GFP_KERNEL);
2587 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2588 if (conn_param->ord > ep->ird) {
2589 ep->ird = conn_param->ird;
2590 ep->ord = conn_param->ord;
2591 send_mpa_reject(ep, conn_param->private_data,
2592 conn_param->private_data_len);
2593 abort_connection(ep, NULL, GFP_KERNEL);
2597 if (conn_param->ird > ep->ord) {
2599 conn_param->ird = 1;
2601 abort_connection(ep, NULL, GFP_KERNEL);
2608 ep->ird = conn_param->ird;
2609 ep->ord = conn_param->ord;
2611 if (ep->mpa_attr.version != 2)
2612 if (peer2peer && ep->ird == 0)
2615 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2617 cm_id->add_ref(cm_id);
2618 ep->com.cm_id = cm_id;
2622 /* bind QP to EP and move to RTS */
2623 attrs.mpa_attr = ep->mpa_attr;
2624 attrs.max_ird = ep->ird;
2625 attrs.max_ord = ep->ord;
2626 attrs.llp_stream_handle = ep;
2627 attrs.next_state = C4IW_QP_STATE_RTS;
2629 /* bind QP and TID with INIT_WR */
2630 mask = C4IW_QP_ATTR_NEXT_STATE |
2631 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2632 C4IW_QP_ATTR_MPA_ATTR |
2633 C4IW_QP_ATTR_MAX_IRD |
2634 C4IW_QP_ATTR_MAX_ORD;
2636 err = c4iw_modify_qp(ep->com.qp->rhp,
2637 ep->com.qp, mask, &attrs, 1);
2640 err = send_mpa_reply(ep, conn_param->private_data,
2641 conn_param->private_data_len);
2645 state_set(&ep->com, FPDU_MODE);
2646 established_upcall(ep);
2647 c4iw_put_ep(&ep->com);
2650 ep->com.cm_id = NULL;
2651 cm_id->rem_ref(cm_id);
2653 c4iw_put_ep(&ep->com);
2657 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2659 struct in_device *ind;
2661 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2662 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2664 ind = in_dev_get(dev->rdev.lldi.ports[0]);
2666 return -EADDRNOTAVAIL;
2667 for_primary_ifa(ind) {
2668 laddr->sin_addr.s_addr = ifa->ifa_address;
2669 raddr->sin_addr.s_addr = ifa->ifa_address;
2675 return found ? 0 : -EADDRNOTAVAIL;
2678 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
2679 unsigned char banned_flags)
2681 struct inet6_dev *idev;
2682 int err = -EADDRNOTAVAIL;
2685 idev = __in6_dev_get(dev);
2687 struct inet6_ifaddr *ifp;
2689 read_lock_bh(&idev->lock);
2690 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2691 if (ifp->scope == IFA_LINK &&
2692 !(ifp->flags & banned_flags)) {
2693 memcpy(addr, &ifp->addr, 16);
2698 read_unlock_bh(&idev->lock);
2704 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
2706 struct in6_addr uninitialized_var(addr);
2707 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2708 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
2710 if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
2711 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
2712 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
2715 return -EADDRNOTAVAIL;
2718 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2720 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2723 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
2724 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
2725 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&cm_id->local_addr;
2726 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2727 &cm_id->remote_addr;
2731 if ((conn_param->ord > c4iw_max_read_depth) ||
2732 (conn_param->ird > c4iw_max_read_depth)) {
2736 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2738 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2742 init_timer(&ep->timer);
2743 ep->plen = conn_param->private_data_len;
2745 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2746 conn_param->private_data, ep->plen);
2747 ep->ird = conn_param->ird;
2748 ep->ord = conn_param->ord;
2750 if (peer2peer && ep->ord == 0)
2753 cm_id->add_ref(cm_id);
2755 ep->com.cm_id = cm_id;
2756 ep->com.qp = get_qhp(dev, conn_param->qpn);
2758 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
2763 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2767 * Allocate an active TID to initiate a TCP connection.
2769 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2770 if (ep->atid == -1) {
2771 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2775 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
2777 if (cm_id->remote_addr.ss_family == AF_INET) {
2779 ra = (__u8 *)&raddr->sin_addr;
2782 * Handle loopback requests to INADDR_ANY.
2784 if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
2785 err = pick_local_ipaddrs(dev, cm_id);
2791 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
2792 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
2793 ra, ntohs(raddr->sin_port));
2794 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
2795 raddr->sin_addr.s_addr, laddr->sin_port,
2796 raddr->sin_port, 0);
2799 ra = (__u8 *)&raddr6->sin6_addr;
2802 * Handle loopback requests to INADDR_ANY.
2804 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
2805 err = pick_local_ip6addrs(dev, cm_id);
2811 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
2812 __func__, laddr6->sin6_addr.s6_addr,
2813 ntohs(laddr6->sin6_port),
2814 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
2815 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
2816 raddr6->sin6_addr.s6_addr,
2817 laddr6->sin6_port, raddr6->sin6_port, 0,
2818 raddr6->sin6_scope_id);
2821 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2822 err = -EHOSTUNREACH;
2826 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
2828 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2832 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2833 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2836 state_set(&ep->com, CONNECTING);
2838 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2839 sizeof(ep->com.local_addr));
2840 memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
2841 sizeof(ep->com.remote_addr));
2843 /* send connect request to rnic */
2844 err = send_connect(ep);
2848 cxgb4_l2t_release(ep->l2t);
2850 dst_release(ep->dst);
2852 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2853 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2855 cm_id->rem_ref(cm_id);
2856 c4iw_put_ep(&ep->com);
2861 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2864 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2866 c4iw_init_wr_wait(&ep->com.wr_wait);
2867 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
2868 ep->stid, &sin6->sin6_addr,
2870 ep->com.dev->rdev.lldi.rxq_ids[0]);
2872 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2876 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
2878 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
2882 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
2885 struct sockaddr_in *sin = (struct sockaddr_in *)&ep->com.local_addr;
2887 if (dev->rdev.lldi.enable_fw_ofld_conn) {
2889 err = cxgb4_create_server_filter(
2890 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2891 sin->sin_addr.s_addr, sin->sin_port, 0,
2892 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
2893 if (err == -EBUSY) {
2894 set_current_state(TASK_UNINTERRUPTIBLE);
2895 schedule_timeout(usecs_to_jiffies(100));
2897 } while (err == -EBUSY);
2899 c4iw_init_wr_wait(&ep->com.wr_wait);
2900 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2901 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
2902 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
2904 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2909 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
2911 &sin->sin_addr, ntohs(sin->sin_port));
2915 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2918 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2919 struct c4iw_listen_ep *ep;
2923 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2925 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2929 PDBG("%s ep %p\n", __func__, ep);
2930 cm_id->add_ref(cm_id);
2931 ep->com.cm_id = cm_id;
2933 ep->backlog = backlog;
2934 memcpy(&ep->com.local_addr, &cm_id->local_addr,
2935 sizeof(ep->com.local_addr));
2938 * Allocate a server TID.
2940 if (dev->rdev.lldi.enable_fw_ofld_conn)
2941 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
2942 cm_id->local_addr.ss_family, ep);
2944 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
2945 cm_id->local_addr.ss_family, ep);
2947 if (ep->stid == -1) {
2948 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2952 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
2953 state_set(&ep->com, LISTEN);
2954 if (ep->com.local_addr.ss_family == AF_INET)
2955 err = create_server4(dev, ep);
2957 err = create_server6(dev, ep);
2959 cm_id->provider_data = ep;
2962 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2963 ep->com.local_addr.ss_family);
2965 cm_id->rem_ref(cm_id);
2966 c4iw_put_ep(&ep->com);
2972 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2975 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2977 PDBG("%s ep %p\n", __func__, ep);
2980 state_set(&ep->com, DEAD);
2981 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
2982 ep->com.local_addr.ss_family == AF_INET) {
2983 err = cxgb4_remove_server_filter(
2984 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2985 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2987 c4iw_init_wr_wait(&ep->com.wr_wait);
2988 err = cxgb4_remove_server(
2989 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2990 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2993 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
2996 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
2997 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
2998 ep->com.local_addr.ss_family);
3000 cm_id->rem_ref(cm_id);
3001 c4iw_put_ep(&ep->com);
3005 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3010 struct c4iw_rdev *rdev;
3012 mutex_lock(&ep->com.mutex);
3014 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3015 states[ep->com.state], abrupt);
3017 rdev = &ep->com.dev->rdev;
3018 if (c4iw_fatal_error(rdev)) {
3020 close_complete_upcall(ep);
3021 ep->com.state = DEAD;
3023 switch (ep->com.state) {
3031 ep->com.state = ABORTING;
3033 ep->com.state = CLOSING;
3036 set_bit(CLOSE_SENT, &ep->com.flags);
3039 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3043 ep->com.state = ABORTING;
3045 ep->com.state = MORIBUND;
3051 PDBG("%s ignoring disconnect ep %p state %u\n",
3052 __func__, ep, ep->com.state);
3061 set_bit(EP_DISC_ABORT, &ep->com.history);
3062 close_complete_upcall(ep);
3063 ret = send_abort(ep, NULL, gfp);
3065 set_bit(EP_DISC_CLOSE, &ep->com.history);
3066 ret = send_halfclose(ep, gfp);
3071 mutex_unlock(&ep->com.mutex);
3073 release_ep_resources(ep);
3077 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3078 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3081 int atid = be32_to_cpu(req->tid);
3083 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3084 (__force u32) req->tid);
3088 switch (req->retval) {
3090 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3091 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3092 send_fw_act_open_req(ep, atid);
3096 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3097 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3098 send_fw_act_open_req(ep, atid);
3103 pr_info("%s unexpected ofld conn wr retval %d\n",
3104 __func__, req->retval);
3107 pr_err("active ofld_connect_wr failure %d atid %d\n",
3109 mutex_lock(&dev->rdev.stats.lock);
3110 dev->rdev.stats.act_ofld_conn_fails++;
3111 mutex_unlock(&dev->rdev.stats.lock);
3112 connect_reply_upcall(ep, status2errno(req->retval));
3113 state_set(&ep->com, DEAD);
3114 remove_handle(dev, &dev->atid_idr, atid);
3115 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3116 dst_release(ep->dst);
3117 cxgb4_l2t_release(ep->l2t);
3118 c4iw_put_ep(&ep->com);
3121 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3122 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3124 struct sk_buff *rpl_skb;
3125 struct cpl_pass_accept_req *cpl;
3128 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3131 PDBG("%s passive open failure %d\n", __func__, req->retval);
3132 mutex_lock(&dev->rdev.stats.lock);
3133 dev->rdev.stats.pas_ofld_conn_fails++;
3134 mutex_unlock(&dev->rdev.stats.lock);
3137 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3138 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3139 (__force u32) htonl(
3140 (__force u32) req->tid)));
3141 ret = pass_accept_req(dev, rpl_skb);
3148 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3150 struct cpl_fw6_msg *rpl = cplhdr(skb);
3151 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3153 switch (rpl->type) {
3155 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3157 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3158 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3159 switch (req->t_state) {
3161 active_ofld_conn_reply(dev, skb, req);
3164 passive_ofld_conn_reply(dev, skb, req);
3167 pr_err("%s unexpected ofld conn wr state %d\n",
3168 __func__, req->t_state);
3176 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3179 u16 vlantag, len, hdr_len, eth_hdr_len;
3181 struct cpl_rx_pkt *cpl = cplhdr(skb);
3182 struct cpl_pass_accept_req *req;
3183 struct tcp_options_received tmp_opt;
3184 struct c4iw_dev *dev;
3186 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3187 /* Store values from cpl_rx_pkt in temporary location. */
3188 vlantag = (__force u16) cpl->vlan;
3189 len = (__force u16) cpl->len;
3190 l2info = (__force u32) cpl->l2info;
3191 hdr_len = (__force u16) cpl->hdr_len;
3194 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3197 * We need to parse the TCP options from SYN packet.
3198 * to generate cpl_pass_accept_req.
3200 memset(&tmp_opt, 0, sizeof(tmp_opt));
3201 tcp_clear_options(&tmp_opt);
3202 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3204 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3205 memset(req, 0, sizeof(*req));
3206 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
3207 V_SYN_MAC_IDX(G_RX_MACIDX(
3208 (__force int) htonl(l2info))) |
3210 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3211 G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
3212 G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
3213 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
3214 (__force int) htonl(l2info))) |
3215 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
3216 (__force int) htons(hdr_len))) |
3217 V_IP_HDR_LEN(G_RX_IPHDR_LEN(
3218 (__force int) htons(hdr_len))) |
3219 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
3220 req->vlan = (__force __be16) vlantag;
3221 req->len = (__force __be16) len;
3222 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
3223 PASS_OPEN_TOS(tos));
3224 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3225 if (tmp_opt.wscale_ok)
3226 req->tcpopt.wsf = tmp_opt.snd_wscale;
3227 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3228 if (tmp_opt.sack_ok)
3229 req->tcpopt.sack = 1;
3230 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3234 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3235 __be32 laddr, __be16 lport,
3236 __be32 raddr, __be16 rport,
3237 u32 rcv_isn, u32 filter, u16 window,
3238 u32 rss_qid, u8 port_id)
3240 struct sk_buff *req_skb;
3241 struct fw_ofld_connection_wr *req;
3242 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3244 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3245 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3246 memset(req, 0, sizeof(*req));
3247 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
3248 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
3249 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
3250 req->le.filter = (__force __be32) filter;
3251 req->le.lport = lport;
3252 req->le.pport = rport;
3253 req->le.u.ipv4.lip = laddr;
3254 req->le.u.ipv4.pip = raddr;
3255 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3256 req->tcb.rcv_adv = htons(window);
3257 req->tcb.t_state_to_astid =
3258 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
3259 V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
3260 V_FW_OFLD_CONNECTION_WR_ASTID(
3261 GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
3264 * We store the qid in opt2 which will be used by the firmware
3265 * to send us the wr response.
3267 req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
3270 * We initialize the MSS index in TCB to 0xF.
3271 * So that when driver sends cpl_pass_accept_rpl
3272 * TCB picks up the correct value. If this was 0
3273 * TP will ignore any value > 0 for MSS index.
3275 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
3276 req->cookie = (unsigned long)skb;
3278 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3279 cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3283 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3284 * messages when a filter is being used instead of server to
3285 * redirect a syn packet. When packets hit filter they are redirected
3286 * to the offload queue and driver tries to establish the connection
3287 * using firmware work request.
3289 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3292 unsigned int filter;
3293 struct ethhdr *eh = NULL;
3294 struct vlan_ethhdr *vlan_eh = NULL;
3296 struct tcphdr *tcph;
3297 struct rss_header *rss = (void *)skb->data;
3298 struct cpl_rx_pkt *cpl = (void *)skb->data;
3299 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3300 struct l2t_entry *e;
3301 struct dst_entry *dst;
3302 struct c4iw_ep *lep;
3304 struct port_info *pi;
3305 struct net_device *pdev;
3306 u16 rss_qid, eth_hdr_len;
3309 struct neighbour *neigh;
3311 /* Drop all non-SYN packets */
3312 if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
3316 * Drop all packets which did not hit the filter.
3317 * Unlikely to happen.
3319 if (!(rss->filter_hit && rss->filter_tid))
3323 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3325 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val)
3326 - dev->rdev.lldi.tids->sftid_base
3327 + dev->rdev.lldi.tids->nstids;
3329 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3331 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3335 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
3336 G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
3337 G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
3338 if (eth_hdr_len == ETH_HLEN) {
3339 eh = (struct ethhdr *)(req + 1);
3340 iph = (struct iphdr *)(eh + 1);
3342 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3343 iph = (struct iphdr *)(vlan_eh + 1);
3344 skb->vlan_tci = ntohs(cpl->vlan);
3347 if (iph->version != 0x4)
3350 tcph = (struct tcphdr *)(iph + 1);
3351 skb_set_network_header(skb, (void *)iph - (void *)rss);
3352 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3355 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3356 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3357 ntohs(tcph->source), iph->tos);
3359 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3362 pr_err("%s - failed to find dst entry!\n",
3366 neigh = dst_neigh_lookup_skb(dst, skb);
3369 pr_err("%s - failed to allocate neigh!\n",
3374 if (neigh->dev->flags & IFF_LOOPBACK) {
3375 pdev = ip_dev_find(&init_net, iph->daddr);
3376 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3378 pi = (struct port_info *)netdev_priv(pdev);
3379 tx_chan = cxgb4_port_chan(pdev);
3382 pdev = get_real_dev(neigh->dev);
3383 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3385 pi = (struct port_info *)netdev_priv(pdev);
3386 tx_chan = cxgb4_port_chan(pdev);
3389 pr_err("%s - failed to allocate l2t entry!\n",
3394 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3395 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3396 window = (__force u16) htons((__force u16)tcph->window);
3398 /* Calcuate filter portion for LE region. */
3399 filter = (__force unsigned int) cpu_to_be32(select_ntuple(dev, dst, e));
3402 * Synthesize the cpl_pass_accept_req. We have everything except the
3403 * TID. Once firmware sends a reply with TID we update the TID field
3404 * in cpl and pass it through the regular cpl_pass_accept_req path.
3406 build_cpl_pass_accept_req(skb, stid, iph->tos);
3407 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3408 tcph->source, ntohl(tcph->seq), filter, window,
3409 rss_qid, pi->port_id);
3410 cxgb4_l2t_release(e);
3418 * These are the real handlers that are called from a
3421 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
3422 [CPL_ACT_ESTABLISH] = act_establish,
3423 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3424 [CPL_RX_DATA] = rx_data,
3425 [CPL_ABORT_RPL_RSS] = abort_rpl,
3426 [CPL_ABORT_RPL] = abort_rpl,
3427 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3428 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3429 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3430 [CPL_PASS_ESTABLISH] = pass_establish,
3431 [CPL_PEER_CLOSE] = peer_close,
3432 [CPL_ABORT_REQ_RSS] = peer_abort,
3433 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3434 [CPL_RDMA_TERMINATE] = terminate,
3435 [CPL_FW4_ACK] = fw4_ack,
3436 [CPL_FW6_MSG] = deferred_fw6_msg,
3437 [CPL_RX_PKT] = rx_pkt
3440 static void process_timeout(struct c4iw_ep *ep)
3442 struct c4iw_qp_attributes attrs;
3445 mutex_lock(&ep->com.mutex);
3446 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3448 set_bit(TIMEDOUT, &ep->com.history);
3449 switch (ep->com.state) {
3451 __state_set(&ep->com, ABORTING);
3452 connect_reply_upcall(ep, -ETIMEDOUT);
3455 __state_set(&ep->com, ABORTING);
3459 if (ep->com.cm_id && ep->com.qp) {
3460 attrs.next_state = C4IW_QP_STATE_ERROR;
3461 c4iw_modify_qp(ep->com.qp->rhp,
3462 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3465 __state_set(&ep->com, ABORTING);
3468 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3469 __func__, ep, ep->hwtid, ep->com.state);
3472 mutex_unlock(&ep->com.mutex);
3474 abort_connection(ep, NULL, GFP_KERNEL);
3475 c4iw_put_ep(&ep->com);
3478 static void process_timedout_eps(void)
3482 spin_lock_irq(&timeout_lock);
3483 while (!list_empty(&timeout_list)) {
3484 struct list_head *tmp;
3486 tmp = timeout_list.next;
3488 spin_unlock_irq(&timeout_lock);
3489 ep = list_entry(tmp, struct c4iw_ep, entry);
3490 process_timeout(ep);
3491 spin_lock_irq(&timeout_lock);
3493 spin_unlock_irq(&timeout_lock);
3496 static void process_work(struct work_struct *work)
3498 struct sk_buff *skb = NULL;
3499 struct c4iw_dev *dev;
3500 struct cpl_act_establish *rpl;
3501 unsigned int opcode;
3504 while ((skb = skb_dequeue(&rxq))) {
3506 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3507 opcode = rpl->ot.opcode;
3509 BUG_ON(!work_handlers[opcode]);
3510 ret = work_handlers[opcode](dev, skb);
3514 process_timedout_eps();
3517 static DECLARE_WORK(skb_work, process_work);
3519 static void ep_timeout(unsigned long arg)
3521 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3524 spin_lock(&timeout_lock);
3525 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3526 list_add_tail(&ep->entry, &timeout_list);
3529 spin_unlock(&timeout_lock);
3531 queue_work(workq, &skb_work);
3535 * All the CM events are handled on a work queue to have a safe context.
3537 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3541 * Save dev in the skb->cb area.
3543 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3546 * Queue the skb and schedule the worker thread.
3548 skb_queue_tail(&rxq, skb);
3549 queue_work(workq, &skb_work);
3553 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3555 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3557 if (rpl->status != CPL_ERR_NONE) {
3558 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3559 "for tid %u\n", rpl->status, GET_TID(rpl));
3565 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3567 struct cpl_fw6_msg *rpl = cplhdr(skb);
3568 struct c4iw_wr_wait *wr_waitp;
3571 PDBG("%s type %u\n", __func__, rpl->type);
3573 switch (rpl->type) {
3574 case FW6_TYPE_WR_RPL:
3575 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3576 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3577 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3579 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3583 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3587 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3595 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3597 struct cpl_abort_req_rss *req = cplhdr(skb);
3599 struct tid_info *t = dev->rdev.lldi.tids;
3600 unsigned int tid = GET_TID(req);
3602 ep = lookup_tid(t, tid);
3604 printk(KERN_WARNING MOD
3605 "Abort on non-existent endpoint, tid %d\n", tid);
3609 if (is_neg_adv_abort(req->status)) {
3610 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3615 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3619 * Wake up any threads in rdma_init() or rdma_fini().
3620 * However, if we are on MPAv2 and want to retry with MPAv1
3621 * then, don't wake up yet.
3623 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3624 if (ep->com.state != MPA_REQ_SENT)
3625 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3627 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3633 * Most upcalls from the T4 Core go to sched() to
3634 * schedule the processing on a work queue.
3636 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3637 [CPL_ACT_ESTABLISH] = sched,
3638 [CPL_ACT_OPEN_RPL] = sched,
3639 [CPL_RX_DATA] = sched,
3640 [CPL_ABORT_RPL_RSS] = sched,
3641 [CPL_ABORT_RPL] = sched,
3642 [CPL_PASS_OPEN_RPL] = sched,
3643 [CPL_CLOSE_LISTSRV_RPL] = sched,
3644 [CPL_PASS_ACCEPT_REQ] = sched,
3645 [CPL_PASS_ESTABLISH] = sched,
3646 [CPL_PEER_CLOSE] = sched,
3647 [CPL_CLOSE_CON_RPL] = sched,
3648 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
3649 [CPL_RDMA_TERMINATE] = sched,
3650 [CPL_FW4_ACK] = sched,
3651 [CPL_SET_TCB_RPL] = set_tcb_rpl,
3652 [CPL_FW6_MSG] = fw6_msg,
3653 [CPL_RX_PKT] = sched
3656 int __init c4iw_cm_init(void)
3658 spin_lock_init(&timeout_lock);
3659 skb_queue_head_init(&rxq);
3661 workq = create_singlethread_workqueue("iw_cxgb4");
3668 void __exit c4iw_cm_term(void)
3670 WARN_ON(!list_empty(&timeout_list));
3671 flush_workqueue(workq);
3672 destroy_workqueue(workq);