2 * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
50 #include <rdma/ib_addr.h>
55 static char *states[] = {
72 module_param(nocong, int, 0644);
73 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
75 static int enable_ecn;
76 module_param(enable_ecn, int, 0644);
77 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
79 static int dack_mode = 1;
80 module_param(dack_mode, int, 0644);
81 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
83 uint c4iw_max_read_depth = 32;
84 module_param(c4iw_max_read_depth, int, 0644);
85 MODULE_PARM_DESC(c4iw_max_read_depth,
86 "Per-connection max ORD/IRD (default=32)");
88 static int enable_tcp_timestamps;
89 module_param(enable_tcp_timestamps, int, 0644);
90 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
92 static int enable_tcp_sack;
93 module_param(enable_tcp_sack, int, 0644);
94 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
96 static int enable_tcp_window_scaling = 1;
97 module_param(enable_tcp_window_scaling, int, 0644);
98 MODULE_PARM_DESC(enable_tcp_window_scaling,
99 "Enable tcp window scaling (default=1)");
102 module_param(c4iw_debug, int, 0644);
103 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
105 static int peer2peer = 1;
106 module_param(peer2peer, int, 0644);
107 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
109 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
110 module_param(p2p_type, int, 0644);
111 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
112 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
114 static int ep_timeout_secs = 60;
115 module_param(ep_timeout_secs, int, 0644);
116 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
117 "in seconds (default=60)");
119 static int mpa_rev = 2;
120 module_param(mpa_rev, int, 0644);
121 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
122 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
123 " compliant (default=2)");
125 static int markers_enabled;
126 module_param(markers_enabled, int, 0644);
127 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
129 static int crc_enabled = 1;
130 module_param(crc_enabled, int, 0644);
131 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
133 static int rcv_win = 256 * 1024;
134 module_param(rcv_win, int, 0644);
135 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
137 static int snd_win = 128 * 1024;
138 module_param(snd_win, int, 0644);
139 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
141 static struct workqueue_struct *workq;
143 static struct sk_buff_head rxq;
145 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
146 static void ep_timeout(unsigned long arg);
147 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
148 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
150 static LIST_HEAD(timeout_list);
151 static spinlock_t timeout_lock;
153 static void deref_cm_id(struct c4iw_ep_common *epc)
155 epc->cm_id->rem_ref(epc->cm_id);
157 set_bit(CM_ID_DEREFED, &epc->history);
160 static void ref_cm_id(struct c4iw_ep_common *epc)
162 set_bit(CM_ID_REFED, &epc->history);
163 epc->cm_id->add_ref(epc->cm_id);
166 static void deref_qp(struct c4iw_ep *ep)
168 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
169 clear_bit(QP_REFERENCED, &ep->com.flags);
170 set_bit(QP_DEREFED, &ep->com.history);
173 static void ref_qp(struct c4iw_ep *ep)
175 set_bit(QP_REFERENCED, &ep->com.flags);
176 set_bit(QP_REFED, &ep->com.history);
177 c4iw_qp_add_ref(&ep->com.qp->ibqp);
180 static void start_ep_timer(struct c4iw_ep *ep)
182 PDBG("%s ep %p\n", __func__, ep);
183 if (timer_pending(&ep->timer)) {
184 pr_err("%s timer already started! ep %p\n",
188 clear_bit(TIMEOUT, &ep->com.flags);
189 c4iw_get_ep(&ep->com);
190 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
191 ep->timer.data = (unsigned long)ep;
192 ep->timer.function = ep_timeout;
193 add_timer(&ep->timer);
196 static int stop_ep_timer(struct c4iw_ep *ep)
198 PDBG("%s ep %p stopping\n", __func__, ep);
199 del_timer_sync(&ep->timer);
200 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
201 c4iw_put_ep(&ep->com);
207 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
208 struct l2t_entry *l2e)
212 if (c4iw_fatal_error(rdev)) {
214 PDBG("%s - device in error state - dropping\n", __func__);
217 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
220 else if (error == NET_XMIT_DROP)
222 return error < 0 ? error : 0;
225 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
229 if (c4iw_fatal_error(rdev)) {
231 PDBG("%s - device in error state - dropping\n", __func__);
234 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
237 return error < 0 ? error : 0;
240 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
242 struct cpl_tid_release *req;
244 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
247 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
248 INIT_TP_WR(req, hwtid);
249 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
250 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
251 c4iw_ofld_send(rdev, skb);
255 static void set_emss(struct c4iw_ep *ep, u16 opt)
257 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
258 ((AF_INET == ep->com.remote_addr.ss_family) ?
259 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
260 sizeof(struct tcphdr);
262 if (TCPOPT_TSTAMP_G(opt))
263 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
267 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
268 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
269 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
273 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
275 enum c4iw_ep_state state;
277 mutex_lock(&epc->mutex);
279 mutex_unlock(&epc->mutex);
283 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
288 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
290 mutex_lock(&epc->mutex);
291 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
292 __state_set(epc, new);
293 mutex_unlock(&epc->mutex);
297 static void *alloc_ep(int size, gfp_t gfp)
299 struct c4iw_ep_common *epc;
301 epc = kzalloc(size, gfp);
303 kref_init(&epc->kref);
304 mutex_init(&epc->mutex);
305 c4iw_init_wr_wait(&epc->wr_wait);
307 PDBG("%s alloc ep %p\n", __func__, epc);
311 static void remove_ep_tid(struct c4iw_ep *ep)
315 spin_lock_irqsave(&ep->com.dev->lock, flags);
316 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
317 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
320 static void insert_ep_tid(struct c4iw_ep *ep)
324 spin_lock_irqsave(&ep->com.dev->lock, flags);
325 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
326 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
330 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
332 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
337 spin_lock_irqsave(&dev->lock, flags);
338 ep = idr_find(&dev->hwtid_idr, tid);
340 c4iw_get_ep(&ep->com);
341 spin_unlock_irqrestore(&dev->lock, flags);
346 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
348 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
351 struct c4iw_listen_ep *ep;
354 spin_lock_irqsave(&dev->lock, flags);
355 ep = idr_find(&dev->stid_idr, stid);
357 c4iw_get_ep(&ep->com);
358 spin_unlock_irqrestore(&dev->lock, flags);
362 void _c4iw_free_ep(struct kref *kref)
366 ep = container_of(kref, struct c4iw_ep, com.kref);
367 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
368 if (test_bit(QP_REFERENCED, &ep->com.flags))
370 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
371 if (ep->com.remote_addr.ss_family == AF_INET6) {
372 struct sockaddr_in6 *sin6 =
373 (struct sockaddr_in6 *)
377 ep->com.dev->rdev.lldi.ports[0],
378 (const u32 *)&sin6->sin6_addr.s6_addr,
381 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
382 dst_release(ep->dst);
383 cxgb4_l2t_release(ep->l2t);
385 kfree_skb(ep->mpa_skb);
390 static void release_ep_resources(struct c4iw_ep *ep)
392 set_bit(RELEASE_RESOURCES, &ep->com.flags);
395 * If we have a hwtid, then remove it from the idr table
396 * so lookups will no longer find this endpoint. Otherwise
397 * we have a race where one thread finds the ep ptr just
398 * before the other thread is freeing the ep memory.
402 c4iw_put_ep(&ep->com);
405 static int status2errno(int status)
410 case CPL_ERR_CONN_RESET:
412 case CPL_ERR_ARP_MISS:
413 return -EHOSTUNREACH;
414 case CPL_ERR_CONN_TIMEDOUT:
416 case CPL_ERR_TCAM_FULL:
418 case CPL_ERR_CONN_EXIST:
426 * Try and reuse skbs already allocated...
428 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
430 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
433 skb_reset_transport_header(skb);
435 skb = alloc_skb(len, gfp);
437 t4_set_arp_err_handler(skb, NULL, NULL);
441 static struct net_device *get_real_dev(struct net_device *egress_dev)
443 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
446 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
450 egress_dev = get_real_dev(egress_dev);
451 for (i = 0; i < dev->rdev.lldi.nports; i++)
452 if (dev->rdev.lldi.ports[i] == egress_dev)
457 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
458 __u8 *peer_ip, __be16 local_port,
459 __be16 peer_port, u8 tos,
462 struct dst_entry *dst = NULL;
464 if (IS_ENABLED(CONFIG_IPV6)) {
467 memset(&fl6, 0, sizeof(fl6));
468 memcpy(&fl6.daddr, peer_ip, 16);
469 memcpy(&fl6.saddr, local_ip, 16);
470 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
471 fl6.flowi6_oif = sin6_scope_id;
472 dst = ip6_route_output(&init_net, NULL, &fl6);
475 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
476 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
486 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
487 __be32 peer_ip, __be16 local_port,
488 __be16 peer_port, u8 tos)
494 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
495 peer_port, local_port, IPPROTO_TCP,
499 n = dst_neigh_lookup(&rt->dst, &peer_ip);
502 if (!our_interface(dev, n->dev) &&
503 !(n->dev->flags & IFF_LOOPBACK)) {
505 dst_release(&rt->dst);
512 static void arp_failure_discard(void *handle, struct sk_buff *skb)
514 pr_err(MOD "ARP failure\n");
518 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
520 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
525 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
526 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
529 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
533 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
534 release_ep_resources(ep);
538 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
542 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
543 c4iw_put_ep(&ep->parent_ep->com);
544 release_ep_resources(ep);
549 * Fake up a special CPL opcode and call sched() so process_work() will call
550 * _put_ep_safe() in a safe context to free the ep resources. This is needed
551 * because ARP error handlers are called in an ATOMIC context, and
552 * _c4iw_free_ep() needs to block.
554 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
557 struct cpl_act_establish *rpl = cplhdr(skb);
559 /* Set our special ARP_FAILURE opcode */
560 rpl->ot.opcode = cpl;
563 * Save ep in the skb->cb area, after where sched() will save the dev
566 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
567 sched(ep->com.dev, skb);
570 /* Handle an ARP failure for an accept */
571 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
573 struct c4iw_ep *ep = handle;
575 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
578 __state_set(&ep->com, DEAD);
579 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
583 * Handle an ARP failure for an active open.
585 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
587 struct c4iw_ep *ep = handle;
589 printk(KERN_ERR MOD "ARP failure during connect\n");
590 connect_reply_upcall(ep, -EHOSTUNREACH);
591 __state_set(&ep->com, DEAD);
592 if (ep->com.remote_addr.ss_family == AF_INET6) {
593 struct sockaddr_in6 *sin6 =
594 (struct sockaddr_in6 *)&ep->com.local_addr;
595 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
596 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
598 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
599 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
600 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
604 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
607 static void abort_arp_failure(void *handle, struct sk_buff *skb)
610 struct c4iw_ep *ep = handle;
611 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
612 struct cpl_abort_req *req = cplhdr(skb);
614 PDBG("%s rdev %p\n", __func__, rdev);
615 req->cmd = CPL_ABORT_NO_RST;
616 ret = c4iw_ofld_send(rdev, skb);
618 __state_set(&ep->com, DEAD);
619 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
623 static int send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
625 unsigned int flowclen = 80;
626 struct fw_flowc_wr *flowc;
628 u16 vlan = ep->l2t->vlan;
631 if (vlan == CPL_L2T_VLAN_NONE)
636 skb = get_skb(skb, flowclen, GFP_KERNEL);
637 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
639 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
640 FW_FLOWC_WR_NPARAMS_V(nparams));
641 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(flowclen,
642 16)) | FW_WR_FLOWID_V(ep->hwtid));
644 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
645 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
646 (ep->com.dev->rdev.lldi.pf));
647 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
648 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
649 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
650 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
651 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
652 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
653 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
654 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
655 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
656 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
657 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
658 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
659 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
660 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
664 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
665 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
666 flowc->mnemval[8].val = cpu_to_be32(pri);
668 /* Pad WR to 16 byte boundary */
669 flowc->mnemval[8].mnemonic = 0;
670 flowc->mnemval[8].val = 0;
672 for (i = 0; i < 9; i++) {
673 flowc->mnemval[i].r4[0] = 0;
674 flowc->mnemval[i].r4[1] = 0;
675 flowc->mnemval[i].r4[2] = 0;
678 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
679 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
682 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
684 struct cpl_close_con_req *req;
686 int wrlen = roundup(sizeof *req, 16);
688 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
689 skb = get_skb(NULL, wrlen, gfp);
691 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
694 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
695 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
696 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
697 memset(req, 0, wrlen);
698 INIT_TP_WR(req, ep->hwtid);
699 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
701 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
704 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
706 struct cpl_abort_req *req;
707 int wrlen = roundup(sizeof *req, 16);
709 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
710 skb = get_skb(skb, wrlen, gfp);
712 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
716 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
717 t4_set_arp_err_handler(skb, ep, abort_arp_failure);
718 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
719 memset(req, 0, wrlen);
720 INIT_TP_WR(req, ep->hwtid);
721 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
722 req->cmd = CPL_ABORT_SEND_RST;
723 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
726 static void best_mtu(const unsigned short *mtus, unsigned short mtu,
727 unsigned int *idx, int use_ts, int ipv6)
729 unsigned short hdr_size = (ipv6 ?
730 sizeof(struct ipv6hdr) :
731 sizeof(struct iphdr)) +
732 sizeof(struct tcphdr) +
734 round_up(TCPOLEN_TIMESTAMP, 4) : 0);
735 unsigned short data_size = mtu - hdr_size;
737 cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
740 static int send_connect(struct c4iw_ep *ep)
742 struct cpl_act_open_req *req = NULL;
743 struct cpl_t5_act_open_req *t5req = NULL;
744 struct cpl_t6_act_open_req *t6req = NULL;
745 struct cpl_act_open_req6 *req6 = NULL;
746 struct cpl_t5_act_open_req6 *t5req6 = NULL;
747 struct cpl_t6_act_open_req6 *t6req6 = NULL;
751 unsigned int mtu_idx;
753 int win, sizev4, sizev6, wrlen;
754 struct sockaddr_in *la = (struct sockaddr_in *)
756 struct sockaddr_in *ra = (struct sockaddr_in *)
757 &ep->com.remote_addr;
758 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
760 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
761 &ep->com.remote_addr;
763 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
764 u32 isn = (prandom_u32() & ~7UL) - 1;
766 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
768 sizev4 = sizeof(struct cpl_act_open_req);
769 sizev6 = sizeof(struct cpl_act_open_req6);
772 sizev4 = sizeof(struct cpl_t5_act_open_req);
773 sizev6 = sizeof(struct cpl_t5_act_open_req6);
776 sizev4 = sizeof(struct cpl_t6_act_open_req);
777 sizev6 = sizeof(struct cpl_t6_act_open_req6);
780 pr_err("T%d Chip is not supported\n",
781 CHELSIO_CHIP_VERSION(adapter_type));
785 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
786 roundup(sizev4, 16) :
789 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
791 skb = get_skb(NULL, wrlen, GFP_KERNEL);
793 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
797 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
799 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
800 enable_tcp_timestamps,
801 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
802 wscale = compute_wscale(rcv_win);
805 * Specify the largest window that will fit in opt0. The
806 * remainder will be specified in the rx_data_ack.
808 win = ep->rcv_win >> 10;
809 if (win > RCV_BUFSIZ_M)
812 opt0 = (nocong ? NO_CONG_F : 0) |
815 WND_SCALE_V(wscale) |
817 L2T_IDX_V(ep->l2t->idx) |
818 TX_CHAN_V(ep->tx_chan) |
819 SMAC_SEL_V(ep->smac_idx) |
820 DSCP_V(ep->tos >> 2) |
821 ULP_MODE_V(ULP_MODE_TCPDDP) |
823 opt2 = RX_CHANNEL_V(0) |
824 CCTRL_ECN_V(enable_ecn) |
825 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
826 if (enable_tcp_timestamps)
827 opt2 |= TSTAMPS_EN_F;
830 if (wscale && enable_tcp_window_scaling)
831 opt2 |= WND_SCALE_EN_F;
832 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
836 opt2 |= T5_OPT_2_VALID_F;
837 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
841 if (ep->com.remote_addr.ss_family == AF_INET6)
842 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
843 (const u32 *)&la6->sin6_addr.s6_addr, 1);
845 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
847 if (ep->com.remote_addr.ss_family == AF_INET) {
848 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
850 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
854 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
856 INIT_TP_WR(t5req, 0);
857 req = (struct cpl_act_open_req *)t5req;
860 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
862 INIT_TP_WR(t6req, 0);
863 req = (struct cpl_act_open_req *)t6req;
864 t5req = (struct cpl_t5_act_open_req *)t6req;
867 pr_err("T%d Chip is not supported\n",
868 CHELSIO_CHIP_VERSION(adapter_type));
873 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
874 ((ep->rss_qid<<14) | ep->atid)));
875 req->local_port = la->sin_port;
876 req->peer_port = ra->sin_port;
877 req->local_ip = la->sin_addr.s_addr;
878 req->peer_ip = ra->sin_addr.s_addr;
879 req->opt0 = cpu_to_be64(opt0);
881 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
882 req->params = cpu_to_be32(cxgb4_select_ntuple(
883 ep->com.dev->rdev.lldi.ports[0],
885 req->opt2 = cpu_to_be32(opt2);
887 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
889 ep->com.dev->rdev.lldi.ports[0],
891 t5req->rsvd = cpu_to_be32(isn);
892 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
893 t5req->opt2 = cpu_to_be32(opt2);
896 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
898 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
902 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
904 INIT_TP_WR(t5req6, 0);
905 req6 = (struct cpl_act_open_req6 *)t5req6;
908 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
910 INIT_TP_WR(t6req6, 0);
911 req6 = (struct cpl_act_open_req6 *)t6req6;
912 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
915 pr_err("T%d Chip is not supported\n",
916 CHELSIO_CHIP_VERSION(adapter_type));
921 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
922 ((ep->rss_qid<<14)|ep->atid)));
923 req6->local_port = la6->sin6_port;
924 req6->peer_port = ra6->sin6_port;
925 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
926 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
927 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
928 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
929 req6->opt0 = cpu_to_be64(opt0);
931 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
932 req6->params = cpu_to_be32(cxgb4_select_ntuple(
933 ep->com.dev->rdev.lldi.ports[0],
935 req6->opt2 = cpu_to_be32(opt2);
937 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
939 ep->com.dev->rdev.lldi.ports[0],
941 t5req6->rsvd = cpu_to_be32(isn);
942 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
943 t5req6->opt2 = cpu_to_be32(opt2);
947 set_bit(ACT_OPEN_REQ, &ep->com.history);
948 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
950 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
951 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
952 (const u32 *)&la6->sin6_addr.s6_addr, 1);
956 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
959 int mpalen, wrlen, ret;
960 struct fw_ofld_tx_data_wr *req;
961 struct mpa_message *mpa;
962 struct mpa_v2_conn_params mpa_v2_params;
964 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
966 BUG_ON(skb_cloned(skb));
968 mpalen = sizeof(*mpa) + ep->plen;
969 if (mpa_rev_to_use == 2)
970 mpalen += sizeof(struct mpa_v2_conn_params);
971 wrlen = roundup(mpalen + sizeof *req, 16);
972 skb = get_skb(skb, wrlen, GFP_KERNEL);
974 connect_reply_upcall(ep, -ENOMEM);
977 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
979 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
980 memset(req, 0, wrlen);
981 req->op_to_immdlen = cpu_to_be32(
982 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
984 FW_WR_IMMDLEN_V(mpalen));
985 req->flowid_len16 = cpu_to_be32(
986 FW_WR_FLOWID_V(ep->hwtid) |
987 FW_WR_LEN16_V(wrlen >> 4));
988 req->plen = cpu_to_be32(mpalen);
989 req->tunnel_to_proxy = cpu_to_be32(
990 FW_OFLD_TX_DATA_WR_FLUSH_F |
991 FW_OFLD_TX_DATA_WR_SHOVE_F);
993 mpa = (struct mpa_message *)(req + 1);
994 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
995 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
996 (markers_enabled ? MPA_MARKERS : 0) |
997 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
998 mpa->private_data_size = htons(ep->plen);
999 mpa->revision = mpa_rev_to_use;
1000 if (mpa_rev_to_use == 1) {
1001 ep->tried_with_mpa_v1 = 1;
1002 ep->retry_with_mpa_v1 = 0;
1005 if (mpa_rev_to_use == 2) {
1006 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1007 sizeof (struct mpa_v2_conn_params));
1008 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1010 mpa_v2_params.ird = htons((u16)ep->ird);
1011 mpa_v2_params.ord = htons((u16)ep->ord);
1014 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1015 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1016 mpa_v2_params.ord |=
1017 htons(MPA_V2_RDMA_WRITE_RTR);
1018 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1019 mpa_v2_params.ord |=
1020 htons(MPA_V2_RDMA_READ_RTR);
1022 memcpy(mpa->private_data, &mpa_v2_params,
1023 sizeof(struct mpa_v2_conn_params));
1026 memcpy(mpa->private_data +
1027 sizeof(struct mpa_v2_conn_params),
1028 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1031 memcpy(mpa->private_data,
1032 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1035 * Reference the mpa skb. This ensures the data area
1036 * will remain in memory until the hw acks the tx.
1037 * Function fw4_ack() will deref it.
1040 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1041 BUG_ON(ep->mpa_skb);
1043 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1047 __state_set(&ep->com, MPA_REQ_SENT);
1048 ep->mpa_attr.initiator = 1;
1049 ep->snd_seq += mpalen;
1053 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1056 struct fw_ofld_tx_data_wr *req;
1057 struct mpa_message *mpa;
1058 struct sk_buff *skb;
1059 struct mpa_v2_conn_params mpa_v2_params;
1061 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1063 mpalen = sizeof(*mpa) + plen;
1064 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1065 mpalen += sizeof(struct mpa_v2_conn_params);
1066 wrlen = roundup(mpalen + sizeof *req, 16);
1068 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1070 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1073 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1075 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1076 memset(req, 0, wrlen);
1077 req->op_to_immdlen = cpu_to_be32(
1078 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1080 FW_WR_IMMDLEN_V(mpalen));
1081 req->flowid_len16 = cpu_to_be32(
1082 FW_WR_FLOWID_V(ep->hwtid) |
1083 FW_WR_LEN16_V(wrlen >> 4));
1084 req->plen = cpu_to_be32(mpalen);
1085 req->tunnel_to_proxy = cpu_to_be32(
1086 FW_OFLD_TX_DATA_WR_FLUSH_F |
1087 FW_OFLD_TX_DATA_WR_SHOVE_F);
1089 mpa = (struct mpa_message *)(req + 1);
1090 memset(mpa, 0, sizeof(*mpa));
1091 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1092 mpa->flags = MPA_REJECT;
1093 mpa->revision = ep->mpa_attr.version;
1094 mpa->private_data_size = htons(plen);
1096 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1097 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1098 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1099 sizeof (struct mpa_v2_conn_params));
1100 mpa_v2_params.ird = htons(((u16)ep->ird) |
1101 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1103 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1105 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1106 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1107 FW_RI_INIT_P2PTYPE_READ_REQ ?
1108 MPA_V2_RDMA_READ_RTR : 0) : 0));
1109 memcpy(mpa->private_data, &mpa_v2_params,
1110 sizeof(struct mpa_v2_conn_params));
1113 memcpy(mpa->private_data +
1114 sizeof(struct mpa_v2_conn_params), pdata, plen);
1117 memcpy(mpa->private_data, pdata, plen);
1120 * Reference the mpa skb again. This ensures the data area
1121 * will remain in memory until the hw acks the tx.
1122 * Function fw4_ack() will deref it.
1125 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1126 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1127 BUG_ON(ep->mpa_skb);
1129 ep->snd_seq += mpalen;
1130 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1133 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1136 struct fw_ofld_tx_data_wr *req;
1137 struct mpa_message *mpa;
1138 struct sk_buff *skb;
1139 struct mpa_v2_conn_params mpa_v2_params;
1141 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1143 mpalen = sizeof(*mpa) + plen;
1144 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1145 mpalen += sizeof(struct mpa_v2_conn_params);
1146 wrlen = roundup(mpalen + sizeof *req, 16);
1148 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1150 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1153 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1155 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1156 memset(req, 0, wrlen);
1157 req->op_to_immdlen = cpu_to_be32(
1158 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1160 FW_WR_IMMDLEN_V(mpalen));
1161 req->flowid_len16 = cpu_to_be32(
1162 FW_WR_FLOWID_V(ep->hwtid) |
1163 FW_WR_LEN16_V(wrlen >> 4));
1164 req->plen = cpu_to_be32(mpalen);
1165 req->tunnel_to_proxy = cpu_to_be32(
1166 FW_OFLD_TX_DATA_WR_FLUSH_F |
1167 FW_OFLD_TX_DATA_WR_SHOVE_F);
1169 mpa = (struct mpa_message *)(req + 1);
1170 memset(mpa, 0, sizeof(*mpa));
1171 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1172 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1173 (markers_enabled ? MPA_MARKERS : 0);
1174 mpa->revision = ep->mpa_attr.version;
1175 mpa->private_data_size = htons(plen);
1177 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1178 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1179 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1180 sizeof (struct mpa_v2_conn_params));
1181 mpa_v2_params.ird = htons((u16)ep->ird);
1182 mpa_v2_params.ord = htons((u16)ep->ord);
1183 if (peer2peer && (ep->mpa_attr.p2p_type !=
1184 FW_RI_INIT_P2PTYPE_DISABLED)) {
1185 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1187 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1188 mpa_v2_params.ord |=
1189 htons(MPA_V2_RDMA_WRITE_RTR);
1190 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1191 mpa_v2_params.ord |=
1192 htons(MPA_V2_RDMA_READ_RTR);
1195 memcpy(mpa->private_data, &mpa_v2_params,
1196 sizeof(struct mpa_v2_conn_params));
1199 memcpy(mpa->private_data +
1200 sizeof(struct mpa_v2_conn_params), pdata, plen);
1203 memcpy(mpa->private_data, pdata, plen);
1206 * Reference the mpa skb. This ensures the data area
1207 * will remain in memory until the hw acks the tx.
1208 * Function fw4_ack() will deref it.
1211 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1213 __state_set(&ep->com, MPA_REP_SENT);
1214 ep->snd_seq += mpalen;
1215 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1218 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1221 struct cpl_act_establish *req = cplhdr(skb);
1222 unsigned int tid = GET_TID(req);
1223 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1224 struct tid_info *t = dev->rdev.lldi.tids;
1227 ep = lookup_atid(t, atid);
1229 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1230 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1232 mutex_lock(&ep->com.mutex);
1233 dst_confirm(ep->dst);
1235 /* setup the hwtid for this connection */
1237 cxgb4_insert_tid(t, ep, tid);
1240 ep->snd_seq = be32_to_cpu(req->snd_isn);
1241 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1243 set_emss(ep, ntohs(req->tcp_opt));
1245 /* dealloc the atid */
1246 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1247 cxgb4_free_atid(t, atid);
1248 set_bit(ACT_ESTAB, &ep->com.history);
1250 /* start MPA negotiation */
1251 ret = send_flowc(ep, NULL);
1254 if (ep->retry_with_mpa_v1)
1255 ret = send_mpa_req(ep, skb, 1);
1257 ret = send_mpa_req(ep, skb, mpa_rev);
1260 mutex_unlock(&ep->com.mutex);
1263 mutex_unlock(&ep->com.mutex);
1264 connect_reply_upcall(ep, -ENOMEM);
1265 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1269 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1271 struct iw_cm_event event;
1273 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1274 memset(&event, 0, sizeof(event));
1275 event.event = IW_CM_EVENT_CLOSE;
1276 event.status = status;
1277 if (ep->com.cm_id) {
1278 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1279 ep, ep->com.cm_id, ep->hwtid);
1280 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1281 deref_cm_id(&ep->com);
1282 set_bit(CLOSE_UPCALL, &ep->com.history);
1286 static void peer_close_upcall(struct c4iw_ep *ep)
1288 struct iw_cm_event event;
1290 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1291 memset(&event, 0, sizeof(event));
1292 event.event = IW_CM_EVENT_DISCONNECT;
1293 if (ep->com.cm_id) {
1294 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1295 ep, ep->com.cm_id, ep->hwtid);
1296 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1297 set_bit(DISCONN_UPCALL, &ep->com.history);
1301 static void peer_abort_upcall(struct c4iw_ep *ep)
1303 struct iw_cm_event event;
1305 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1306 memset(&event, 0, sizeof(event));
1307 event.event = IW_CM_EVENT_CLOSE;
1308 event.status = -ECONNRESET;
1309 if (ep->com.cm_id) {
1310 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1311 ep->com.cm_id, ep->hwtid);
1312 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1313 deref_cm_id(&ep->com);
1314 set_bit(ABORT_UPCALL, &ep->com.history);
1318 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1320 struct iw_cm_event event;
1322 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1323 memset(&event, 0, sizeof(event));
1324 event.event = IW_CM_EVENT_CONNECT_REPLY;
1325 event.status = status;
1326 memcpy(&event.local_addr, &ep->com.local_addr,
1327 sizeof(ep->com.local_addr));
1328 memcpy(&event.remote_addr, &ep->com.remote_addr,
1329 sizeof(ep->com.remote_addr));
1331 if ((status == 0) || (status == -ECONNREFUSED)) {
1332 if (!ep->tried_with_mpa_v1) {
1333 /* this means MPA_v2 is used */
1334 event.ord = ep->ird;
1335 event.ird = ep->ord;
1336 event.private_data_len = ep->plen -
1337 sizeof(struct mpa_v2_conn_params);
1338 event.private_data = ep->mpa_pkt +
1339 sizeof(struct mpa_message) +
1340 sizeof(struct mpa_v2_conn_params);
1342 /* this means MPA_v1 is used */
1343 event.ord = cur_max_read_depth(ep->com.dev);
1344 event.ird = cur_max_read_depth(ep->com.dev);
1345 event.private_data_len = ep->plen;
1346 event.private_data = ep->mpa_pkt +
1347 sizeof(struct mpa_message);
1351 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1353 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1354 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1357 deref_cm_id(&ep->com);
1360 static int connect_request_upcall(struct c4iw_ep *ep)
1362 struct iw_cm_event event;
1365 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1366 memset(&event, 0, sizeof(event));
1367 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1368 memcpy(&event.local_addr, &ep->com.local_addr,
1369 sizeof(ep->com.local_addr));
1370 memcpy(&event.remote_addr, &ep->com.remote_addr,
1371 sizeof(ep->com.remote_addr));
1372 event.provider_data = ep;
1373 if (!ep->tried_with_mpa_v1) {
1374 /* this means MPA_v2 is used */
1375 event.ord = ep->ord;
1376 event.ird = ep->ird;
1377 event.private_data_len = ep->plen -
1378 sizeof(struct mpa_v2_conn_params);
1379 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1380 sizeof(struct mpa_v2_conn_params);
1382 /* this means MPA_v1 is used. Send max supported */
1383 event.ord = cur_max_read_depth(ep->com.dev);
1384 event.ird = cur_max_read_depth(ep->com.dev);
1385 event.private_data_len = ep->plen;
1386 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1388 c4iw_get_ep(&ep->com);
1389 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1392 c4iw_put_ep(&ep->com);
1393 set_bit(CONNREQ_UPCALL, &ep->com.history);
1394 c4iw_put_ep(&ep->parent_ep->com);
1398 static void established_upcall(struct c4iw_ep *ep)
1400 struct iw_cm_event event;
1402 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1403 memset(&event, 0, sizeof(event));
1404 event.event = IW_CM_EVENT_ESTABLISHED;
1405 event.ird = ep->ord;
1406 event.ord = ep->ird;
1407 if (ep->com.cm_id) {
1408 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1409 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1410 set_bit(ESTAB_UPCALL, &ep->com.history);
1414 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1416 struct cpl_rx_data_ack *req;
1417 struct sk_buff *skb;
1418 int wrlen = roundup(sizeof *req, 16);
1420 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1421 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1423 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1428 * If we couldn't specify the entire rcv window at connection setup
1429 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1430 * then add the overage in to the credits returned.
1432 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1433 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1435 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1436 memset(req, 0, wrlen);
1437 INIT_TP_WR(req, ep->hwtid);
1438 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1440 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
1442 RX_DACK_MODE_V(dack_mode));
1443 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1444 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1448 #define RELAXED_IRD_NEGOTIATION 1
1451 * process_mpa_reply - process streaming mode MPA reply
1455 * 0 upon success indicating a connect request was delivered to the ULP
1456 * or the mpa request is incomplete but valid so far.
1458 * 1 if a failure requires the caller to close the connection.
1460 * 2 if a failure requires the caller to abort the connection.
1462 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1464 struct mpa_message *mpa;
1465 struct mpa_v2_conn_params *mpa_v2_params;
1467 u16 resp_ird, resp_ord;
1468 u8 rtr_mismatch = 0, insuff_ird = 0;
1469 struct c4iw_qp_attributes attrs;
1470 enum c4iw_qp_attr_mask mask;
1474 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1477 * If we get more than the supported amount of private data
1478 * then we must fail this connection.
1480 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1482 goto err_stop_timer;
1486 * copy the new data into our accumulation buffer.
1488 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1490 ep->mpa_pkt_len += skb->len;
1493 * if we don't even have the mpa message, then bail.
1495 if (ep->mpa_pkt_len < sizeof(*mpa))
1497 mpa = (struct mpa_message *) ep->mpa_pkt;
1499 /* Validate MPA header. */
1500 if (mpa->revision > mpa_rev) {
1501 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1502 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1504 goto err_stop_timer;
1506 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1508 goto err_stop_timer;
1511 plen = ntohs(mpa->private_data_size);
1514 * Fail if there's too much private data.
1516 if (plen > MPA_MAX_PRIVATE_DATA) {
1518 goto err_stop_timer;
1522 * If plen does not account for pkt size
1524 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1526 goto err_stop_timer;
1529 ep->plen = (u8) plen;
1532 * If we don't have all the pdata yet, then bail.
1533 * We'll continue process when more data arrives.
1535 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1538 if (mpa->flags & MPA_REJECT) {
1539 err = -ECONNREFUSED;
1540 goto err_stop_timer;
1544 * Stop mpa timer. If it expired, then
1545 * we ignore the MPA reply. process_timeout()
1546 * will abort the connection.
1548 if (stop_ep_timer(ep))
1552 * If we get here we have accumulated the entire mpa
1553 * start reply message including private data. And
1554 * the MPA header is valid.
1556 __state_set(&ep->com, FPDU_MODE);
1557 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1558 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1559 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1560 ep->mpa_attr.version = mpa->revision;
1561 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1563 if (mpa->revision == 2) {
1564 ep->mpa_attr.enhanced_rdma_conn =
1565 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1566 if (ep->mpa_attr.enhanced_rdma_conn) {
1567 mpa_v2_params = (struct mpa_v2_conn_params *)
1568 (ep->mpa_pkt + sizeof(*mpa));
1569 resp_ird = ntohs(mpa_v2_params->ird) &
1570 MPA_V2_IRD_ORD_MASK;
1571 resp_ord = ntohs(mpa_v2_params->ord) &
1572 MPA_V2_IRD_ORD_MASK;
1573 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1574 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1577 * This is a double-check. Ideally, below checks are
1578 * not required since ird/ord stuff has been taken
1579 * care of in c4iw_accept_cr
1581 if (ep->ird < resp_ord) {
1582 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1583 ep->com.dev->rdev.lldi.max_ordird_qp)
1587 } else if (ep->ird > resp_ord) {
1590 if (ep->ord > resp_ird) {
1591 if (RELAXED_IRD_NEGOTIATION)
1602 if (ntohs(mpa_v2_params->ird) &
1603 MPA_V2_PEER2PEER_MODEL) {
1604 if (ntohs(mpa_v2_params->ord) &
1605 MPA_V2_RDMA_WRITE_RTR)
1606 ep->mpa_attr.p2p_type =
1607 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1608 else if (ntohs(mpa_v2_params->ord) &
1609 MPA_V2_RDMA_READ_RTR)
1610 ep->mpa_attr.p2p_type =
1611 FW_RI_INIT_P2PTYPE_READ_REQ;
1614 } else if (mpa->revision == 1)
1616 ep->mpa_attr.p2p_type = p2p_type;
1618 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1619 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1620 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1621 ep->mpa_attr.recv_marker_enabled,
1622 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1623 ep->mpa_attr.p2p_type, p2p_type);
1626 * If responder's RTR does not match with that of initiator, assign
1627 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1628 * generated when moving QP to RTS state.
1629 * A TERM message will be sent after QP has moved to RTS state
1631 if ((ep->mpa_attr.version == 2) && peer2peer &&
1632 (ep->mpa_attr.p2p_type != p2p_type)) {
1633 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1637 attrs.mpa_attr = ep->mpa_attr;
1638 attrs.max_ird = ep->ird;
1639 attrs.max_ord = ep->ord;
1640 attrs.llp_stream_handle = ep;
1641 attrs.next_state = C4IW_QP_STATE_RTS;
1643 mask = C4IW_QP_ATTR_NEXT_STATE |
1644 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1645 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1647 /* bind QP and TID with INIT_WR */
1648 err = c4iw_modify_qp(ep->com.qp->rhp,
1649 ep->com.qp, mask, &attrs, 1);
1654 * If responder's RTR requirement did not match with what initiator
1655 * supports, generate TERM message
1658 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1659 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1660 attrs.ecode = MPA_NOMATCH_RTR;
1661 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1662 attrs.send_term = 1;
1663 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1664 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1671 * Generate TERM if initiator IRD is not sufficient for responder
1672 * provided ORD. Currently, we do the same behaviour even when
1673 * responder provided IRD is also not sufficient as regards to
1677 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1679 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1680 attrs.ecode = MPA_INSUFF_IRD;
1681 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1682 attrs.send_term = 1;
1683 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1684 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1695 connect_reply_upcall(ep, err);
1700 * process_mpa_request - process streaming mode MPA request
1704 * 0 upon success indicating a connect request was delivered to the ULP
1705 * or the mpa request is incomplete but valid so far.
1707 * 1 if a failure requires the caller to close the connection.
1709 * 2 if a failure requires the caller to abort the connection.
1711 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1713 struct mpa_message *mpa;
1714 struct mpa_v2_conn_params *mpa_v2_params;
1717 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1720 * If we get more than the supported amount of private data
1721 * then we must fail this connection.
1723 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1724 goto err_stop_timer;
1726 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1729 * Copy the new data into our accumulation buffer.
1731 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1733 ep->mpa_pkt_len += skb->len;
1736 * If we don't even have the mpa message, then bail.
1737 * We'll continue process when more data arrives.
1739 if (ep->mpa_pkt_len < sizeof(*mpa))
1742 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1743 mpa = (struct mpa_message *) ep->mpa_pkt;
1746 * Validate MPA Header.
1748 if (mpa->revision > mpa_rev) {
1749 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1750 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1751 goto err_stop_timer;
1754 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1755 goto err_stop_timer;
1757 plen = ntohs(mpa->private_data_size);
1760 * Fail if there's too much private data.
1762 if (plen > MPA_MAX_PRIVATE_DATA)
1763 goto err_stop_timer;
1766 * If plen does not account for pkt size
1768 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1769 goto err_stop_timer;
1770 ep->plen = (u8) plen;
1773 * If we don't have all the pdata yet, then bail.
1775 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1779 * If we get here we have accumulated the entire mpa
1780 * start reply message including private data.
1782 ep->mpa_attr.initiator = 0;
1783 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1784 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1785 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1786 ep->mpa_attr.version = mpa->revision;
1787 if (mpa->revision == 1)
1788 ep->tried_with_mpa_v1 = 1;
1789 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1791 if (mpa->revision == 2) {
1792 ep->mpa_attr.enhanced_rdma_conn =
1793 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1794 if (ep->mpa_attr.enhanced_rdma_conn) {
1795 mpa_v2_params = (struct mpa_v2_conn_params *)
1796 (ep->mpa_pkt + sizeof(*mpa));
1797 ep->ird = ntohs(mpa_v2_params->ird) &
1798 MPA_V2_IRD_ORD_MASK;
1799 ep->ord = ntohs(mpa_v2_params->ord) &
1800 MPA_V2_IRD_ORD_MASK;
1801 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1803 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1805 if (ntohs(mpa_v2_params->ord) &
1806 MPA_V2_RDMA_WRITE_RTR)
1807 ep->mpa_attr.p2p_type =
1808 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1809 else if (ntohs(mpa_v2_params->ord) &
1810 MPA_V2_RDMA_READ_RTR)
1811 ep->mpa_attr.p2p_type =
1812 FW_RI_INIT_P2PTYPE_READ_REQ;
1815 } else if (mpa->revision == 1)
1817 ep->mpa_attr.p2p_type = p2p_type;
1819 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1820 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1821 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1822 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1823 ep->mpa_attr.p2p_type);
1825 __state_set(&ep->com, MPA_REQ_RCVD);
1828 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1829 if (ep->parent_ep->com.state != DEAD) {
1830 if (connect_request_upcall(ep))
1831 goto err_unlock_parent;
1833 goto err_unlock_parent;
1835 mutex_unlock(&ep->parent_ep->com.mutex);
1839 mutex_unlock(&ep->parent_ep->com.mutex);
1842 (void)stop_ep_timer(ep);
1847 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1850 struct cpl_rx_data *hdr = cplhdr(skb);
1851 unsigned int dlen = ntohs(hdr->len);
1852 unsigned int tid = GET_TID(hdr);
1853 __u8 status = hdr->status;
1856 ep = get_ep_from_tid(dev, tid);
1859 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1860 skb_pull(skb, sizeof(*hdr));
1861 skb_trim(skb, dlen);
1862 mutex_lock(&ep->com.mutex);
1864 /* update RX credits */
1865 update_rx_credits(ep, dlen);
1867 switch (ep->com.state) {
1869 ep->rcv_seq += dlen;
1870 disconnect = process_mpa_reply(ep, skb);
1873 ep->rcv_seq += dlen;
1874 disconnect = process_mpa_request(ep, skb);
1877 struct c4iw_qp_attributes attrs;
1878 BUG_ON(!ep->com.qp);
1880 pr_err("%s Unexpected streaming data." \
1881 " qpid %u ep %p state %d tid %u status %d\n",
1882 __func__, ep->com.qp->wq.sq.qid, ep,
1883 ep->com.state, ep->hwtid, status);
1884 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1885 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1886 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1893 mutex_unlock(&ep->com.mutex);
1895 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1896 c4iw_put_ep(&ep->com);
1900 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1903 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1905 unsigned int tid = GET_TID(rpl);
1907 ep = get_ep_from_tid(dev, tid);
1909 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1912 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1913 mutex_lock(&ep->com.mutex);
1914 switch (ep->com.state) {
1916 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1917 __state_set(&ep->com, DEAD);
1921 printk(KERN_ERR "%s ep %p state %d\n",
1922 __func__, ep, ep->com.state);
1925 mutex_unlock(&ep->com.mutex);
1928 release_ep_resources(ep);
1929 c4iw_put_ep(&ep->com);
1933 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1935 struct sk_buff *skb;
1936 struct fw_ofld_connection_wr *req;
1937 unsigned int mtu_idx;
1939 struct sockaddr_in *sin;
1942 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1943 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1944 memset(req, 0, sizeof(*req));
1945 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1946 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1947 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1948 ep->com.dev->rdev.lldi.ports[0],
1950 sin = (struct sockaddr_in *)&ep->com.local_addr;
1951 req->le.lport = sin->sin_port;
1952 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1953 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1954 req->le.pport = sin->sin_port;
1955 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1956 req->tcb.t_state_to_astid =
1957 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1958 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1959 req->tcb.cplrxdataack_cplpassacceptrpl =
1960 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1961 req->tcb.tx_max = (__force __be32) jiffies;
1962 req->tcb.rcv_adv = htons(1);
1963 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1964 enable_tcp_timestamps,
1965 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
1966 wscale = compute_wscale(rcv_win);
1969 * Specify the largest window that will fit in opt0. The
1970 * remainder will be specified in the rx_data_ack.
1972 win = ep->rcv_win >> 10;
1973 if (win > RCV_BUFSIZ_M)
1976 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1977 (nocong ? NO_CONG_F : 0) |
1980 WND_SCALE_V(wscale) |
1981 MSS_IDX_V(mtu_idx) |
1982 L2T_IDX_V(ep->l2t->idx) |
1983 TX_CHAN_V(ep->tx_chan) |
1984 SMAC_SEL_V(ep->smac_idx) |
1985 DSCP_V(ep->tos >> 2) |
1986 ULP_MODE_V(ULP_MODE_TCPDDP) |
1988 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1989 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1991 CCTRL_ECN_V(enable_ecn) |
1992 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1993 if (enable_tcp_timestamps)
1994 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1995 if (enable_tcp_sack)
1996 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1997 if (wscale && enable_tcp_window_scaling)
1998 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1999 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2000 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2001 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2002 set_bit(ACT_OFLD_CONN, &ep->com.history);
2003 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2007 * Return whether a failed active open has allocated a TID
2009 static inline int act_open_has_tid(int status)
2011 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
2012 status != CPL_ERR_ARP_MISS;
2015 /* Returns whether a CPL status conveys negative advice.
2017 static int is_neg_adv(unsigned int status)
2019 return status == CPL_ERR_RTX_NEG_ADVICE ||
2020 status == CPL_ERR_PERSIST_NEG_ADVICE ||
2021 status == CPL_ERR_KEEPALV_NEG_ADVICE;
2024 static char *neg_adv_str(unsigned int status)
2027 case CPL_ERR_RTX_NEG_ADVICE:
2028 return "Retransmit timeout";
2029 case CPL_ERR_PERSIST_NEG_ADVICE:
2030 return "Persist timeout";
2031 case CPL_ERR_KEEPALV_NEG_ADVICE:
2032 return "Keepalive timeout";
2038 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2040 ep->snd_win = snd_win;
2041 ep->rcv_win = rcv_win;
2042 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
2045 #define ACT_OPEN_RETRY_COUNT 2
2047 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2048 struct dst_entry *dst, struct c4iw_dev *cdev,
2049 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2051 struct neighbour *n;
2053 struct net_device *pdev;
2055 n = dst_neigh_lookup(dst, peer_ip);
2061 if (n->dev->flags & IFF_LOOPBACK) {
2063 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2064 else if (IS_ENABLED(CONFIG_IPV6))
2065 for_each_netdev(&init_net, pdev) {
2066 if (ipv6_chk_addr(&init_net,
2067 (struct in6_addr *)peer_ip,
2078 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2079 n, pdev, rt_tos2priority(tos));
2082 ep->mtu = pdev->mtu;
2083 ep->tx_chan = cxgb4_port_chan(pdev);
2084 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2085 cxgb4_port_viid(pdev));
2086 step = cdev->rdev.lldi.ntxq /
2087 cdev->rdev.lldi.nchan;
2088 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2089 step = cdev->rdev.lldi.nrxq /
2090 cdev->rdev.lldi.nchan;
2091 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2092 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2093 cxgb4_port_idx(pdev) * step];
2094 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2097 pdev = get_real_dev(n->dev);
2098 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2102 ep->mtu = dst_mtu(dst);
2103 ep->tx_chan = cxgb4_port_chan(pdev);
2104 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2105 cxgb4_port_viid(pdev));
2106 step = cdev->rdev.lldi.ntxq /
2107 cdev->rdev.lldi.nchan;
2108 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2109 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2110 step = cdev->rdev.lldi.nrxq /
2111 cdev->rdev.lldi.nchan;
2112 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2113 cxgb4_port_idx(pdev) * step];
2114 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2117 ep->retry_with_mpa_v1 = 0;
2118 ep->tried_with_mpa_v1 = 0;
2130 static int c4iw_reconnect(struct c4iw_ep *ep)
2133 struct sockaddr_in *laddr = (struct sockaddr_in *)
2134 &ep->com.cm_id->m_local_addr;
2135 struct sockaddr_in *raddr = (struct sockaddr_in *)
2136 &ep->com.cm_id->m_remote_addr;
2137 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2138 &ep->com.cm_id->m_local_addr;
2139 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2140 &ep->com.cm_id->m_remote_addr;
2144 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2145 init_timer(&ep->timer);
2146 c4iw_init_wr_wait(&ep->com.wr_wait);
2149 * Allocate an active TID to initiate a TCP connection.
2151 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2152 if (ep->atid == -1) {
2153 pr_err("%s - cannot alloc atid.\n", __func__);
2157 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2160 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2161 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
2162 raddr->sin_addr.s_addr, laddr->sin_port,
2163 raddr->sin_port, ep->com.cm_id->tos);
2165 ra = (__u8 *)&raddr->sin_addr;
2167 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
2168 raddr6->sin6_addr.s6_addr,
2169 laddr6->sin6_port, raddr6->sin6_port, 0,
2170 raddr6->sin6_scope_id);
2172 ra = (__u8 *)&raddr6->sin6_addr;
2175 pr_err("%s - cannot find route.\n", __func__);
2176 err = -EHOSTUNREACH;
2179 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2180 ep->com.dev->rdev.lldi.adapter_type,
2181 ep->com.cm_id->tos);
2183 pr_err("%s - cannot alloc l2e.\n", __func__);
2187 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2188 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2191 state_set(&ep->com, CONNECTING);
2192 ep->tos = ep->com.cm_id->tos;
2194 /* send connect request to rnic */
2195 err = send_connect(ep);
2199 cxgb4_l2t_release(ep->l2t);
2201 dst_release(ep->dst);
2203 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2204 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2207 * remember to send notification to upper layer.
2208 * We are in here so the upper layer is not aware that this is
2209 * re-connect attempt and so, upper layer is still waiting for
2210 * response of 1st connect request.
2212 connect_reply_upcall(ep, -ECONNRESET);
2213 c4iw_put_ep(&ep->com);
2218 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2221 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2222 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2223 ntohl(rpl->atid_status)));
2224 struct tid_info *t = dev->rdev.lldi.tids;
2225 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2226 struct sockaddr_in *la;
2227 struct sockaddr_in *ra;
2228 struct sockaddr_in6 *la6;
2229 struct sockaddr_in6 *ra6;
2232 ep = lookup_atid(t, atid);
2233 la = (struct sockaddr_in *)&ep->com.local_addr;
2234 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2235 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2236 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2238 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2239 status, status2errno(status));
2241 if (is_neg_adv(status)) {
2242 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2243 __func__, atid, status, neg_adv_str(status));
2244 ep->stats.connect_neg_adv++;
2245 mutex_lock(&dev->rdev.stats.lock);
2246 dev->rdev.stats.neg_adv++;
2247 mutex_unlock(&dev->rdev.stats.lock);
2251 set_bit(ACT_OPEN_RPL, &ep->com.history);
2254 * Log interesting failures.
2257 case CPL_ERR_CONN_RESET:
2258 case CPL_ERR_CONN_TIMEDOUT:
2260 case CPL_ERR_TCAM_FULL:
2261 mutex_lock(&dev->rdev.stats.lock);
2262 dev->rdev.stats.tcam_full++;
2263 mutex_unlock(&dev->rdev.stats.lock);
2264 if (ep->com.local_addr.ss_family == AF_INET &&
2265 dev->rdev.lldi.enable_fw_ofld_conn) {
2266 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2267 ntohl(rpl->atid_status))));
2273 case CPL_ERR_CONN_EXIST:
2274 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2275 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2276 if (ep->com.remote_addr.ss_family == AF_INET6) {
2277 struct sockaddr_in6 *sin6 =
2278 (struct sockaddr_in6 *)
2279 &ep->com.local_addr;
2281 ep->com.dev->rdev.lldi.ports[0],
2283 &sin6->sin6_addr.s6_addr, 1);
2285 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2287 cxgb4_free_atid(t, atid);
2288 dst_release(ep->dst);
2289 cxgb4_l2t_release(ep->l2t);
2295 if (ep->com.local_addr.ss_family == AF_INET) {
2296 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2297 atid, status, status2errno(status),
2298 &la->sin_addr.s_addr, ntohs(la->sin_port),
2299 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2301 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2302 atid, status, status2errno(status),
2303 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2304 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2310 connect_reply_upcall(ep, status2errno(status));
2311 state_set(&ep->com, DEAD);
2313 if (ep->com.remote_addr.ss_family == AF_INET6) {
2314 struct sockaddr_in6 *sin6 =
2315 (struct sockaddr_in6 *)&ep->com.local_addr;
2316 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2317 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2319 if (status && act_open_has_tid(status))
2320 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2322 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2323 cxgb4_free_atid(t, atid);
2324 dst_release(ep->dst);
2325 cxgb4_l2t_release(ep->l2t);
2326 c4iw_put_ep(&ep->com);
2331 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2333 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2334 unsigned int stid = GET_TID(rpl);
2335 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2338 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2341 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2342 rpl->status, status2errno(rpl->status));
2343 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2344 c4iw_put_ep(&ep->com);
2349 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2351 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2352 unsigned int stid = GET_TID(rpl);
2353 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2355 PDBG("%s ep %p\n", __func__, ep);
2356 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2357 c4iw_put_ep(&ep->com);
2361 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2362 struct cpl_pass_accept_req *req)
2364 struct cpl_pass_accept_rpl *rpl;
2365 unsigned int mtu_idx;
2369 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2371 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2373 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2374 BUG_ON(skb_cloned(skb));
2378 if (!is_t4(adapter_type)) {
2379 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2381 INIT_TP_WR(rpl5, ep->hwtid);
2383 skb_trim(skb, sizeof(*rpl));
2384 INIT_TP_WR(rpl, ep->hwtid);
2386 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2389 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2390 enable_tcp_timestamps && req->tcpopt.tstamp,
2391 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2392 wscale = compute_wscale(rcv_win);
2395 * Specify the largest window that will fit in opt0. The
2396 * remainder will be specified in the rx_data_ack.
2398 win = ep->rcv_win >> 10;
2399 if (win > RCV_BUFSIZ_M)
2401 opt0 = (nocong ? NO_CONG_F : 0) |
2404 WND_SCALE_V(wscale) |
2405 MSS_IDX_V(mtu_idx) |
2406 L2T_IDX_V(ep->l2t->idx) |
2407 TX_CHAN_V(ep->tx_chan) |
2408 SMAC_SEL_V(ep->smac_idx) |
2409 DSCP_V(ep->tos >> 2) |
2410 ULP_MODE_V(ULP_MODE_TCPDDP) |
2412 opt2 = RX_CHANNEL_V(0) |
2413 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2415 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2416 opt2 |= TSTAMPS_EN_F;
2417 if (enable_tcp_sack && req->tcpopt.sack)
2419 if (wscale && enable_tcp_window_scaling)
2420 opt2 |= WND_SCALE_EN_F;
2422 const struct tcphdr *tcph;
2423 u32 hlen = ntohl(req->hdr_len);
2425 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2426 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2429 tcph = (const void *)(req + 1) +
2430 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2431 if (tcph->ece && tcph->cwr)
2432 opt2 |= CCTRL_ECN_V(1);
2434 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2435 u32 isn = (prandom_u32() & ~7UL) - 1;
2436 opt2 |= T5_OPT_2_VALID_F;
2437 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2440 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2443 rpl5->iss = cpu_to_be32(isn);
2444 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2447 rpl->opt0 = cpu_to_be64(opt0);
2448 rpl->opt2 = cpu_to_be32(opt2);
2449 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2450 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2452 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2455 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2457 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2458 BUG_ON(skb_cloned(skb));
2459 skb_trim(skb, sizeof(struct cpl_tid_release));
2460 release_tid(&dev->rdev, hwtid, skb);
2464 static void get_4tuple(struct cpl_pass_accept_req *req, enum chip_type type,
2465 int *iptype, __u8 *local_ip, __u8 *peer_ip,
2466 __be16 *local_port, __be16 *peer_port)
2468 int eth_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2469 ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2470 T6_ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2471 int ip_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2472 IP_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2473 T6_IP_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2474 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2475 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2476 struct tcphdr *tcp = (struct tcphdr *)
2477 ((u8 *)(req + 1) + eth_len + ip_len);
2479 if (ip->version == 4) {
2480 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2481 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2484 memcpy(peer_ip, &ip->saddr, 4);
2485 memcpy(local_ip, &ip->daddr, 4);
2487 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2488 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2491 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2492 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2494 *peer_port = tcp->source;
2495 *local_port = tcp->dest;
2500 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2502 struct c4iw_ep *child_ep = NULL, *parent_ep;
2503 struct cpl_pass_accept_req *req = cplhdr(skb);
2504 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2505 struct tid_info *t = dev->rdev.lldi.tids;
2506 unsigned int hwtid = GET_TID(req);
2507 struct dst_entry *dst;
2508 __u8 local_ip[16], peer_ip[16];
2509 __be16 local_port, peer_port;
2510 struct sockaddr_in6 *sin6;
2512 u16 peer_mss = ntohs(req->tcpopt.mss);
2514 unsigned short hdrs;
2515 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2517 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2519 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2523 if (state_read(&parent_ep->com) != LISTEN) {
2524 PDBG("%s - listening ep not in LISTEN\n", __func__);
2528 get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, &iptype,
2529 local_ip, peer_ip, &local_port, &peer_port);
2531 /* Find output route */
2533 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2534 , __func__, parent_ep, hwtid,
2535 local_ip, peer_ip, ntohs(local_port),
2536 ntohs(peer_port), peer_mss);
2537 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2538 local_port, peer_port,
2541 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2542 , __func__, parent_ep, hwtid,
2543 local_ip, peer_ip, ntohs(local_port),
2544 ntohs(peer_port), peer_mss);
2545 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2546 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2547 ((struct sockaddr_in6 *)
2548 &parent_ep->com.local_addr)->sin6_scope_id);
2551 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2556 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2558 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2564 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2565 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2567 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2574 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2575 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2576 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2577 child_ep->mtu = peer_mss + hdrs;
2579 state_set(&child_ep->com, CONNECTING);
2580 child_ep->com.dev = dev;
2581 child_ep->com.cm_id = NULL;
2584 struct sockaddr_in *sin = (struct sockaddr_in *)
2585 &child_ep->com.local_addr;
2587 sin->sin_family = PF_INET;
2588 sin->sin_port = local_port;
2589 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2591 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2592 sin->sin_family = PF_INET;
2593 sin->sin_port = ((struct sockaddr_in *)
2594 &parent_ep->com.local_addr)->sin_port;
2595 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2597 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2598 sin->sin_family = PF_INET;
2599 sin->sin_port = peer_port;
2600 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2602 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2603 sin6->sin6_family = PF_INET6;
2604 sin6->sin6_port = local_port;
2605 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2607 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2608 sin6->sin6_family = PF_INET6;
2609 sin6->sin6_port = ((struct sockaddr_in6 *)
2610 &parent_ep->com.local_addr)->sin6_port;
2611 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2613 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2614 sin6->sin6_family = PF_INET6;
2615 sin6->sin6_port = peer_port;
2616 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2619 c4iw_get_ep(&parent_ep->com);
2620 child_ep->parent_ep = parent_ep;
2621 child_ep->tos = tos;
2622 child_ep->dst = dst;
2623 child_ep->hwtid = hwtid;
2625 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2626 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2628 init_timer(&child_ep->timer);
2629 cxgb4_insert_tid(t, child_ep, hwtid);
2630 insert_ep_tid(child_ep);
2631 if (accept_cr(child_ep, skb, req)) {
2632 c4iw_put_ep(&parent_ep->com);
2633 release_ep_resources(child_ep);
2635 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2638 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2639 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2640 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2644 reject_cr(dev, hwtid, skb);
2646 c4iw_put_ep(&parent_ep->com);
2651 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2654 struct cpl_pass_establish *req = cplhdr(skb);
2655 unsigned int tid = GET_TID(req);
2658 ep = get_ep_from_tid(dev, tid);
2659 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2660 ep->snd_seq = be32_to_cpu(req->snd_isn);
2661 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2663 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2664 ntohs(req->tcp_opt));
2666 set_emss(ep, ntohs(req->tcp_opt));
2668 dst_confirm(ep->dst);
2669 mutex_lock(&ep->com.mutex);
2670 ep->com.state = MPA_REQ_WAIT;
2672 set_bit(PASS_ESTAB, &ep->com.history);
2673 ret = send_flowc(ep, skb);
2674 mutex_unlock(&ep->com.mutex);
2676 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2677 c4iw_put_ep(&ep->com);
2682 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2684 struct cpl_peer_close *hdr = cplhdr(skb);
2686 struct c4iw_qp_attributes attrs;
2689 unsigned int tid = GET_TID(hdr);
2692 ep = get_ep_from_tid(dev, tid);
2696 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2697 dst_confirm(ep->dst);
2699 set_bit(PEER_CLOSE, &ep->com.history);
2700 mutex_lock(&ep->com.mutex);
2701 switch (ep->com.state) {
2703 __state_set(&ep->com, CLOSING);
2706 __state_set(&ep->com, CLOSING);
2707 connect_reply_upcall(ep, -ECONNRESET);
2712 * We're gonna mark this puppy DEAD, but keep
2713 * the reference on it until the ULP accepts or
2714 * rejects the CR. Also wake up anyone waiting
2715 * in rdma connection migration (see c4iw_accept_cr()).
2717 __state_set(&ep->com, CLOSING);
2718 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2719 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2722 __state_set(&ep->com, CLOSING);
2723 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2724 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2728 __state_set(&ep->com, CLOSING);
2729 attrs.next_state = C4IW_QP_STATE_CLOSING;
2730 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2731 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2732 if (ret != -ECONNRESET) {
2733 peer_close_upcall(ep);
2741 __state_set(&ep->com, MORIBUND);
2745 (void)stop_ep_timer(ep);
2746 if (ep->com.cm_id && ep->com.qp) {
2747 attrs.next_state = C4IW_QP_STATE_IDLE;
2748 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2749 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2751 close_complete_upcall(ep, 0);
2752 __state_set(&ep->com, DEAD);
2762 mutex_unlock(&ep->com.mutex);
2764 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2766 release_ep_resources(ep);
2767 c4iw_put_ep(&ep->com);
2771 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2773 struct cpl_abort_req_rss *req = cplhdr(skb);
2775 struct cpl_abort_rpl *rpl;
2776 struct sk_buff *rpl_skb;
2777 struct c4iw_qp_attributes attrs;
2780 unsigned int tid = GET_TID(req);
2782 ep = get_ep_from_tid(dev, tid);
2786 if (is_neg_adv(req->status)) {
2787 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2788 __func__, ep->hwtid, req->status,
2789 neg_adv_str(req->status));
2790 ep->stats.abort_neg_adv++;
2791 mutex_lock(&dev->rdev.stats.lock);
2792 dev->rdev.stats.neg_adv++;
2793 mutex_unlock(&dev->rdev.stats.lock);
2796 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2798 set_bit(PEER_ABORT, &ep->com.history);
2801 * Wake up any threads in rdma_init() or rdma_fini().
2802 * However, this is not needed if com state is just
2805 if (ep->com.state != MPA_REQ_SENT)
2806 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2808 mutex_lock(&ep->com.mutex);
2809 switch (ep->com.state) {
2811 c4iw_put_ep(&ep->parent_ep->com);
2814 (void)stop_ep_timer(ep);
2817 (void)stop_ep_timer(ep);
2818 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2819 connect_reply_upcall(ep, -ECONNRESET);
2822 * we just don't send notification upwards because we
2823 * want to retry with mpa_v1 without upper layers even
2826 * do some housekeeping so as to re-initiate the
2829 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2831 ep->retry_with_mpa_v1 = 1;
2843 if (ep->com.cm_id && ep->com.qp) {
2844 attrs.next_state = C4IW_QP_STATE_ERROR;
2845 ret = c4iw_modify_qp(ep->com.qp->rhp,
2846 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2850 "%s - qp <- error failed!\n",
2853 peer_abort_upcall(ep);
2858 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2859 mutex_unlock(&ep->com.mutex);
2865 dst_confirm(ep->dst);
2866 if (ep->com.state != ABORTING) {
2867 __state_set(&ep->com, DEAD);
2868 /* we don't release if we want to retry with mpa_v1 */
2869 if (!ep->retry_with_mpa_v1)
2872 mutex_unlock(&ep->com.mutex);
2874 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2876 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2881 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2882 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2883 INIT_TP_WR(rpl, ep->hwtid);
2884 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2885 rpl->cmd = CPL_ABORT_NO_RST;
2886 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2889 release_ep_resources(ep);
2890 else if (ep->retry_with_mpa_v1) {
2891 if (ep->com.remote_addr.ss_family == AF_INET6) {
2892 struct sockaddr_in6 *sin6 =
2893 (struct sockaddr_in6 *)
2894 &ep->com.local_addr;
2896 ep->com.dev->rdev.lldi.ports[0],
2897 (const u32 *)&sin6->sin6_addr.s6_addr,
2900 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2901 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2902 dst_release(ep->dst);
2903 cxgb4_l2t_release(ep->l2t);
2908 c4iw_put_ep(&ep->com);
2909 /* Dereferencing ep, referenced in peer_abort_intr() */
2910 c4iw_put_ep(&ep->com);
2914 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2917 struct c4iw_qp_attributes attrs;
2918 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2920 unsigned int tid = GET_TID(rpl);
2922 ep = get_ep_from_tid(dev, tid);
2926 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2929 /* The cm_id may be null if we failed to connect */
2930 mutex_lock(&ep->com.mutex);
2931 set_bit(CLOSE_CON_RPL, &ep->com.history);
2932 switch (ep->com.state) {
2934 __state_set(&ep->com, MORIBUND);
2937 (void)stop_ep_timer(ep);
2938 if ((ep->com.cm_id) && (ep->com.qp)) {
2939 attrs.next_state = C4IW_QP_STATE_IDLE;
2940 c4iw_modify_qp(ep->com.qp->rhp,
2942 C4IW_QP_ATTR_NEXT_STATE,
2945 close_complete_upcall(ep, 0);
2946 __state_set(&ep->com, DEAD);
2956 mutex_unlock(&ep->com.mutex);
2958 release_ep_resources(ep);
2959 c4iw_put_ep(&ep->com);
2963 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2965 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2966 unsigned int tid = GET_TID(rpl);
2968 struct c4iw_qp_attributes attrs;
2970 ep = get_ep_from_tid(dev, tid);
2973 if (ep && ep->com.qp) {
2974 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2975 ep->com.qp->wq.sq.qid);
2976 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2977 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2978 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2980 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2981 c4iw_put_ep(&ep->com);
2987 * Upcall from the adapter indicating data has been transmitted.
2988 * For us its just the single MPA request or reply. We can now free
2989 * the skb holding the mpa message.
2991 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2994 struct cpl_fw4_ack *hdr = cplhdr(skb);
2995 u8 credits = hdr->credits;
2996 unsigned int tid = GET_TID(hdr);
2999 ep = get_ep_from_tid(dev, tid);
3002 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
3004 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
3005 __func__, ep, ep->hwtid, state_read(&ep->com));
3009 dst_confirm(ep->dst);
3011 PDBG("%s last streaming msg ack ep %p tid %u state %u "
3012 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
3013 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
3014 kfree_skb(ep->mpa_skb);
3016 mutex_lock(&ep->com.mutex);
3017 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3019 mutex_unlock(&ep->com.mutex);
3022 c4iw_put_ep(&ep->com);
3026 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3030 struct c4iw_ep *ep = to_ep(cm_id);
3031 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3033 mutex_lock(&ep->com.mutex);
3034 if (ep->com.state != MPA_REQ_RCVD) {
3035 mutex_unlock(&ep->com.mutex);
3036 c4iw_put_ep(&ep->com);
3039 set_bit(ULP_REJECT, &ep->com.history);
3043 err = send_mpa_reject(ep, pdata, pdata_len);
3046 mutex_unlock(&ep->com.mutex);
3049 err = c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
3051 c4iw_put_ep(&ep->com);
3055 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3058 struct c4iw_qp_attributes attrs;
3059 enum c4iw_qp_attr_mask mask;
3060 struct c4iw_ep *ep = to_ep(cm_id);
3061 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3062 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3065 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3067 mutex_lock(&ep->com.mutex);
3068 if (ep->com.state != MPA_REQ_RCVD) {
3075 set_bit(ULP_ACCEPT, &ep->com.history);
3076 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3077 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3082 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3083 if (conn_param->ord > ep->ird) {
3084 if (RELAXED_IRD_NEGOTIATION) {
3087 ep->ird = conn_param->ird;
3088 ep->ord = conn_param->ord;
3089 send_mpa_reject(ep, conn_param->private_data,
3090 conn_param->private_data_len);
3095 if (conn_param->ird < ep->ord) {
3096 if (RELAXED_IRD_NEGOTIATION &&
3097 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3098 conn_param->ird = ep->ord;
3105 ep->ird = conn_param->ird;
3106 ep->ord = conn_param->ord;
3108 if (ep->mpa_attr.version == 1) {
3109 if (peer2peer && ep->ird == 0)
3113 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3114 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3118 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3120 ep->com.cm_id = cm_id;
3121 ref_cm_id(&ep->com);
3125 /* bind QP to EP and move to RTS */
3126 attrs.mpa_attr = ep->mpa_attr;
3127 attrs.max_ird = ep->ird;
3128 attrs.max_ord = ep->ord;
3129 attrs.llp_stream_handle = ep;
3130 attrs.next_state = C4IW_QP_STATE_RTS;
3132 /* bind QP and TID with INIT_WR */
3133 mask = C4IW_QP_ATTR_NEXT_STATE |
3134 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3135 C4IW_QP_ATTR_MPA_ATTR |
3136 C4IW_QP_ATTR_MAX_IRD |
3137 C4IW_QP_ATTR_MAX_ORD;
3139 err = c4iw_modify_qp(ep->com.qp->rhp,
3140 ep->com.qp, mask, &attrs, 1);
3142 goto err_deref_cm_id;
3144 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3145 err = send_mpa_reply(ep, conn_param->private_data,
3146 conn_param->private_data_len);
3148 goto err_deref_cm_id;
3150 __state_set(&ep->com, FPDU_MODE);
3151 established_upcall(ep);
3152 mutex_unlock(&ep->com.mutex);
3153 c4iw_put_ep(&ep->com);
3156 deref_cm_id(&ep->com);
3160 mutex_unlock(&ep->com.mutex);
3162 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3163 c4iw_put_ep(&ep->com);
3167 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3169 struct in_device *ind;
3171 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3172 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3174 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3176 return -EADDRNOTAVAIL;
3177 for_primary_ifa(ind) {
3178 laddr->sin_addr.s_addr = ifa->ifa_address;
3179 raddr->sin_addr.s_addr = ifa->ifa_address;
3185 return found ? 0 : -EADDRNOTAVAIL;
3188 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3189 unsigned char banned_flags)
3191 struct inet6_dev *idev;
3192 int err = -EADDRNOTAVAIL;
3195 idev = __in6_dev_get(dev);
3197 struct inet6_ifaddr *ifp;
3199 read_lock_bh(&idev->lock);
3200 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3201 if (ifp->scope == IFA_LINK &&
3202 !(ifp->flags & banned_flags)) {
3203 memcpy(addr, &ifp->addr, 16);
3208 read_unlock_bh(&idev->lock);
3214 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3216 struct in6_addr uninitialized_var(addr);
3217 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3218 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3220 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3221 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3222 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3225 return -EADDRNOTAVAIL;
3228 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3230 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3233 struct sockaddr_in *laddr;
3234 struct sockaddr_in *raddr;
3235 struct sockaddr_in6 *laddr6;
3236 struct sockaddr_in6 *raddr6;
3240 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3241 (conn_param->ird > cur_max_read_depth(dev))) {
3245 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3247 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3251 init_timer(&ep->timer);
3252 ep->plen = conn_param->private_data_len;
3254 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3255 conn_param->private_data, ep->plen);
3256 ep->ird = conn_param->ird;
3257 ep->ord = conn_param->ord;
3259 if (peer2peer && ep->ord == 0)
3262 ep->com.cm_id = cm_id;
3263 ref_cm_id(&ep->com);
3265 ep->com.qp = get_qhp(dev, conn_param->qpn);
3267 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3272 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3276 * Allocate an active TID to initiate a TCP connection.
3278 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3279 if (ep->atid == -1) {
3280 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3284 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3286 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3287 sizeof(ep->com.local_addr));
3288 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3289 sizeof(ep->com.remote_addr));
3291 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3292 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3293 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3294 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3296 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3298 ra = (__u8 *)&raddr->sin_addr;
3301 * Handle loopback requests to INADDR_ANY.
3303 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3304 err = pick_local_ipaddrs(dev, cm_id);
3310 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3311 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3312 ra, ntohs(raddr->sin_port));
3313 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
3314 raddr->sin_addr.s_addr, laddr->sin_port,
3315 raddr->sin_port, cm_id->tos);
3318 ra = (__u8 *)&raddr6->sin6_addr;
3321 * Handle loopback requests to INADDR_ANY.
3323 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3324 err = pick_local_ip6addrs(dev, cm_id);
3330 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3331 __func__, laddr6->sin6_addr.s6_addr,
3332 ntohs(laddr6->sin6_port),
3333 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3334 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
3335 raddr6->sin6_addr.s6_addr,
3336 laddr6->sin6_port, raddr6->sin6_port, 0,
3337 raddr6->sin6_scope_id);
3340 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3341 err = -EHOSTUNREACH;
3345 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3346 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3348 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3352 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3353 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3356 state_set(&ep->com, CONNECTING);
3357 ep->tos = cm_id->tos;
3359 /* send connect request to rnic */
3360 err = send_connect(ep);
3364 cxgb4_l2t_release(ep->l2t);
3366 dst_release(ep->dst);
3368 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3369 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3371 deref_cm_id(&ep->com);
3372 c4iw_put_ep(&ep->com);
3377 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3380 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3381 &ep->com.local_addr;
3383 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3384 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3385 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3389 c4iw_init_wr_wait(&ep->com.wr_wait);
3390 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3391 ep->stid, &sin6->sin6_addr,
3393 ep->com.dev->rdev.lldi.rxq_ids[0]);
3395 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3399 err = net_xmit_errno(err);
3401 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3402 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3403 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3405 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3410 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3413 struct sockaddr_in *sin = (struct sockaddr_in *)
3414 &ep->com.local_addr;
3416 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3418 err = cxgb4_create_server_filter(
3419 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3420 sin->sin_addr.s_addr, sin->sin_port, 0,
3421 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3422 if (err == -EBUSY) {
3423 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3427 set_current_state(TASK_UNINTERRUPTIBLE);
3428 schedule_timeout(usecs_to_jiffies(100));
3430 } while (err == -EBUSY);
3432 c4iw_init_wr_wait(&ep->com.wr_wait);
3433 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3434 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3435 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3437 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3441 err = net_xmit_errno(err);
3444 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3446 &sin->sin_addr, ntohs(sin->sin_port));
3450 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3453 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3454 struct c4iw_listen_ep *ep;
3458 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3460 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3464 PDBG("%s ep %p\n", __func__, ep);
3465 ep->com.cm_id = cm_id;
3466 ref_cm_id(&ep->com);
3468 ep->backlog = backlog;
3469 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3470 sizeof(ep->com.local_addr));
3473 * Allocate a server TID.
3475 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3476 ep->com.local_addr.ss_family == AF_INET)
3477 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3478 cm_id->m_local_addr.ss_family, ep);
3480 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3481 cm_id->m_local_addr.ss_family, ep);
3483 if (ep->stid == -1) {
3484 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3488 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3490 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3491 sizeof(ep->com.local_addr));
3493 state_set(&ep->com, LISTEN);
3494 if (ep->com.local_addr.ss_family == AF_INET)
3495 err = create_server4(dev, ep);
3497 err = create_server6(dev, ep);
3499 cm_id->provider_data = ep;
3503 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3504 ep->com.local_addr.ss_family);
3506 deref_cm_id(&ep->com);
3507 c4iw_put_ep(&ep->com);
3513 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3516 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3518 PDBG("%s ep %p\n", __func__, ep);
3521 state_set(&ep->com, DEAD);
3522 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3523 ep->com.local_addr.ss_family == AF_INET) {
3524 err = cxgb4_remove_server_filter(
3525 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3526 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3528 struct sockaddr_in6 *sin6;
3529 c4iw_init_wr_wait(&ep->com.wr_wait);
3530 err = cxgb4_remove_server(
3531 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3532 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3535 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3537 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3538 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3539 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3541 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3542 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3543 ep->com.local_addr.ss_family);
3545 deref_cm_id(&ep->com);
3546 c4iw_put_ep(&ep->com);
3550 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3555 struct c4iw_rdev *rdev;
3557 mutex_lock(&ep->com.mutex);
3559 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3560 states[ep->com.state], abrupt);
3563 * Ref the ep here in case we have fatal errors causing the
3564 * ep to be released and freed.
3566 c4iw_get_ep(&ep->com);
3568 rdev = &ep->com.dev->rdev;
3569 if (c4iw_fatal_error(rdev)) {
3571 close_complete_upcall(ep, -EIO);
3572 ep->com.state = DEAD;
3574 switch (ep->com.state) {
3582 ep->com.state = ABORTING;
3584 ep->com.state = CLOSING;
3587 set_bit(CLOSE_SENT, &ep->com.flags);
3590 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3593 (void)stop_ep_timer(ep);
3594 ep->com.state = ABORTING;
3596 ep->com.state = MORIBUND;
3602 PDBG("%s ignoring disconnect ep %p state %u\n",
3603 __func__, ep, ep->com.state);
3612 set_bit(EP_DISC_ABORT, &ep->com.history);
3613 close_complete_upcall(ep, -ECONNRESET);
3614 ret = send_abort(ep, NULL, gfp);
3616 set_bit(EP_DISC_CLOSE, &ep->com.history);
3617 ret = send_halfclose(ep, gfp);
3620 set_bit(EP_DISC_FAIL, &ep->com.history);
3623 close_complete_upcall(ep, -EIO);
3626 struct c4iw_qp_attributes attrs;
3628 attrs.next_state = C4IW_QP_STATE_ERROR;
3629 ret = c4iw_modify_qp(ep->com.qp->rhp,
3631 C4IW_QP_ATTR_NEXT_STATE,
3635 "%s - qp <- error failed!\n",
3641 mutex_unlock(&ep->com.mutex);
3642 c4iw_put_ep(&ep->com);
3644 release_ep_resources(ep);
3648 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3649 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3652 int atid = be32_to_cpu(req->tid);
3654 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3655 (__force u32) req->tid);
3659 switch (req->retval) {
3661 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3662 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3663 send_fw_act_open_req(ep, atid);
3667 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3668 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3669 send_fw_act_open_req(ep, atid);
3674 pr_info("%s unexpected ofld conn wr retval %d\n",
3675 __func__, req->retval);
3678 pr_err("active ofld_connect_wr failure %d atid %d\n",
3680 mutex_lock(&dev->rdev.stats.lock);
3681 dev->rdev.stats.act_ofld_conn_fails++;
3682 mutex_unlock(&dev->rdev.stats.lock);
3683 connect_reply_upcall(ep, status2errno(req->retval));
3684 state_set(&ep->com, DEAD);
3685 if (ep->com.remote_addr.ss_family == AF_INET6) {
3686 struct sockaddr_in6 *sin6 =
3687 (struct sockaddr_in6 *)&ep->com.local_addr;
3688 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3689 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3691 remove_handle(dev, &dev->atid_idr, atid);
3692 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3693 dst_release(ep->dst);
3694 cxgb4_l2t_release(ep->l2t);
3695 c4iw_put_ep(&ep->com);
3698 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3699 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3701 struct sk_buff *rpl_skb;
3702 struct cpl_pass_accept_req *cpl;
3705 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3708 PDBG("%s passive open failure %d\n", __func__, req->retval);
3709 mutex_lock(&dev->rdev.stats.lock);
3710 dev->rdev.stats.pas_ofld_conn_fails++;
3711 mutex_unlock(&dev->rdev.stats.lock);
3714 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3715 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3716 (__force u32) htonl(
3717 (__force u32) req->tid)));
3718 ret = pass_accept_req(dev, rpl_skb);
3725 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3727 struct cpl_fw6_msg *rpl = cplhdr(skb);
3728 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3730 switch (rpl->type) {
3732 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3734 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3735 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3736 switch (req->t_state) {
3738 active_ofld_conn_reply(dev, skb, req);
3741 passive_ofld_conn_reply(dev, skb, req);
3744 pr_err("%s unexpected ofld conn wr state %d\n",
3745 __func__, req->t_state);
3753 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3756 __be16 hdr_len, vlantag, len;
3758 int tcp_hdr_len, ip_hdr_len;
3760 struct cpl_rx_pkt *cpl = cplhdr(skb);
3761 struct cpl_pass_accept_req *req;
3762 struct tcp_options_received tmp_opt;
3763 struct c4iw_dev *dev;
3764 enum chip_type type;
3766 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3767 /* Store values from cpl_rx_pkt in temporary location. */
3768 vlantag = cpl->vlan;
3770 l2info = cpl->l2info;
3771 hdr_len = cpl->hdr_len;
3774 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3777 * We need to parse the TCP options from SYN packet.
3778 * to generate cpl_pass_accept_req.
3780 memset(&tmp_opt, 0, sizeof(tmp_opt));
3781 tcp_clear_options(&tmp_opt);
3782 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3784 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3785 memset(req, 0, sizeof(*req));
3786 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3787 SYN_MAC_IDX_V(RX_MACIDX_G(
3788 be32_to_cpu(l2info))) |
3790 type = dev->rdev.lldi.adapter_type;
3791 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3792 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3794 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3795 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3796 eth_hdr_len = is_t4(type) ?
3797 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3798 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3799 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3800 IP_HDR_LEN_V(ip_hdr_len) |
3801 ETH_HDR_LEN_V(eth_hdr_len));
3802 } else { /* T6 and later */
3803 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3804 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3805 T6_IP_HDR_LEN_V(ip_hdr_len) |
3806 T6_ETH_HDR_LEN_V(eth_hdr_len));
3808 req->vlan = vlantag;
3810 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3811 PASS_OPEN_TOS_V(tos));
3812 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3813 if (tmp_opt.wscale_ok)
3814 req->tcpopt.wsf = tmp_opt.snd_wscale;
3815 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3816 if (tmp_opt.sack_ok)
3817 req->tcpopt.sack = 1;
3818 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3822 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3823 __be32 laddr, __be16 lport,
3824 __be32 raddr, __be16 rport,
3825 u32 rcv_isn, u32 filter, u16 window,
3826 u32 rss_qid, u8 port_id)
3828 struct sk_buff *req_skb;
3829 struct fw_ofld_connection_wr *req;
3830 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3833 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3834 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3835 memset(req, 0, sizeof(*req));
3836 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3837 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3838 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3839 req->le.filter = (__force __be32) filter;
3840 req->le.lport = lport;
3841 req->le.pport = rport;
3842 req->le.u.ipv4.lip = laddr;
3843 req->le.u.ipv4.pip = raddr;
3844 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3845 req->tcb.rcv_adv = htons(window);
3846 req->tcb.t_state_to_astid =
3847 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3848 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3849 FW_OFLD_CONNECTION_WR_ASTID_V(
3850 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3853 * We store the qid in opt2 which will be used by the firmware
3854 * to send us the wr response.
3856 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3859 * We initialize the MSS index in TCB to 0xF.
3860 * So that when driver sends cpl_pass_accept_rpl
3861 * TCB picks up the correct value. If this was 0
3862 * TP will ignore any value > 0 for MSS index.
3864 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3865 req->cookie = (uintptr_t)skb;
3867 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3868 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3870 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3878 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3879 * messages when a filter is being used instead of server to
3880 * redirect a syn packet. When packets hit filter they are redirected
3881 * to the offload queue and driver tries to establish the connection
3882 * using firmware work request.
3884 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3887 unsigned int filter;
3888 struct ethhdr *eh = NULL;
3889 struct vlan_ethhdr *vlan_eh = NULL;
3891 struct tcphdr *tcph;
3892 struct rss_header *rss = (void *)skb->data;
3893 struct cpl_rx_pkt *cpl = (void *)skb->data;
3894 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3895 struct l2t_entry *e;
3896 struct dst_entry *dst;
3897 struct c4iw_ep *lep = NULL;
3899 struct port_info *pi;
3900 struct net_device *pdev;
3901 u16 rss_qid, eth_hdr_len;
3904 struct neighbour *neigh;
3906 /* Drop all non-SYN packets */
3907 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3911 * Drop all packets which did not hit the filter.
3912 * Unlikely to happen.
3914 if (!(rss->filter_hit && rss->filter_tid))
3918 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3920 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3922 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3924 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3928 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3930 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3933 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3936 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3939 pr_err("T%d Chip is not supported\n",
3940 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3944 if (eth_hdr_len == ETH_HLEN) {
3945 eh = (struct ethhdr *)(req + 1);
3946 iph = (struct iphdr *)(eh + 1);
3948 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3949 iph = (struct iphdr *)(vlan_eh + 1);
3950 skb->vlan_tci = ntohs(cpl->vlan);
3953 if (iph->version != 0x4)
3956 tcph = (struct tcphdr *)(iph + 1);
3957 skb_set_network_header(skb, (void *)iph - (void *)rss);
3958 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3961 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3962 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3963 ntohs(tcph->source), iph->tos);
3965 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3968 pr_err("%s - failed to find dst entry!\n",
3972 neigh = dst_neigh_lookup_skb(dst, skb);
3975 pr_err("%s - failed to allocate neigh!\n",
3980 if (neigh->dev->flags & IFF_LOOPBACK) {
3981 pdev = ip_dev_find(&init_net, iph->daddr);
3982 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3984 pi = (struct port_info *)netdev_priv(pdev);
3985 tx_chan = cxgb4_port_chan(pdev);
3988 pdev = get_real_dev(neigh->dev);
3989 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3991 pi = (struct port_info *)netdev_priv(pdev);
3992 tx_chan = cxgb4_port_chan(pdev);
3994 neigh_release(neigh);
3996 pr_err("%s - failed to allocate l2t entry!\n",
4001 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4002 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4003 window = (__force u16) htons((__force u16)tcph->window);
4005 /* Calcuate filter portion for LE region. */
4006 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4007 dev->rdev.lldi.ports[0],
4011 * Synthesize the cpl_pass_accept_req. We have everything except the
4012 * TID. Once firmware sends a reply with TID we update the TID field
4013 * in cpl and pass it through the regular cpl_pass_accept_req path.
4015 build_cpl_pass_accept_req(skb, stid, iph->tos);
4016 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4017 tcph->source, ntohl(tcph->seq), filter, window,
4018 rss_qid, pi->port_id);
4019 cxgb4_l2t_release(e);
4024 c4iw_put_ep(&lep->com);
4029 * These are the real handlers that are called from a
4032 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4033 [CPL_ACT_ESTABLISH] = act_establish,
4034 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4035 [CPL_RX_DATA] = rx_data,
4036 [CPL_ABORT_RPL_RSS] = abort_rpl,
4037 [CPL_ABORT_RPL] = abort_rpl,
4038 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4039 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4040 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4041 [CPL_PASS_ESTABLISH] = pass_establish,
4042 [CPL_PEER_CLOSE] = peer_close,
4043 [CPL_ABORT_REQ_RSS] = peer_abort,
4044 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4045 [CPL_RDMA_TERMINATE] = terminate,
4046 [CPL_FW4_ACK] = fw4_ack,
4047 [CPL_FW6_MSG] = deferred_fw6_msg,
4048 [CPL_RX_PKT] = rx_pkt,
4049 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4050 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4053 static void process_timeout(struct c4iw_ep *ep)
4055 struct c4iw_qp_attributes attrs;
4058 mutex_lock(&ep->com.mutex);
4059 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4061 set_bit(TIMEDOUT, &ep->com.history);
4062 switch (ep->com.state) {
4064 connect_reply_upcall(ep, -ETIMEDOUT);
4073 if (ep->com.cm_id && ep->com.qp) {
4074 attrs.next_state = C4IW_QP_STATE_ERROR;
4075 c4iw_modify_qp(ep->com.qp->rhp,
4076 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4079 close_complete_upcall(ep, -ETIMEDOUT);
4085 * These states are expected if the ep timed out at the same
4086 * time as another thread was calling stop_ep_timer().
4087 * So we silently do nothing for these states.
4092 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4093 __func__, ep, ep->hwtid, ep->com.state);
4096 mutex_unlock(&ep->com.mutex);
4098 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4099 c4iw_put_ep(&ep->com);
4102 static void process_timedout_eps(void)
4106 spin_lock_irq(&timeout_lock);
4107 while (!list_empty(&timeout_list)) {
4108 struct list_head *tmp;
4110 tmp = timeout_list.next;
4114 spin_unlock_irq(&timeout_lock);
4115 ep = list_entry(tmp, struct c4iw_ep, entry);
4116 process_timeout(ep);
4117 spin_lock_irq(&timeout_lock);
4119 spin_unlock_irq(&timeout_lock);
4122 static void process_work(struct work_struct *work)
4124 struct sk_buff *skb = NULL;
4125 struct c4iw_dev *dev;
4126 struct cpl_act_establish *rpl;
4127 unsigned int opcode;
4130 process_timedout_eps();
4131 while ((skb = skb_dequeue(&rxq))) {
4133 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4134 opcode = rpl->ot.opcode;
4136 BUG_ON(!work_handlers[opcode]);
4137 ret = work_handlers[opcode](dev, skb);
4140 process_timedout_eps();
4144 static DECLARE_WORK(skb_work, process_work);
4146 static void ep_timeout(unsigned long arg)
4148 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4151 spin_lock(&timeout_lock);
4152 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4154 * Only insert if it is not already on the list.
4156 if (!ep->entry.next) {
4157 list_add_tail(&ep->entry, &timeout_list);
4161 spin_unlock(&timeout_lock);
4163 queue_work(workq, &skb_work);
4167 * All the CM events are handled on a work queue to have a safe context.
4169 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4173 * Save dev in the skb->cb area.
4175 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4178 * Queue the skb and schedule the worker thread.
4180 skb_queue_tail(&rxq, skb);
4181 queue_work(workq, &skb_work);
4185 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4187 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4189 if (rpl->status != CPL_ERR_NONE) {
4190 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4191 "for tid %u\n", rpl->status, GET_TID(rpl));
4197 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4199 struct cpl_fw6_msg *rpl = cplhdr(skb);
4200 struct c4iw_wr_wait *wr_waitp;
4203 PDBG("%s type %u\n", __func__, rpl->type);
4205 switch (rpl->type) {
4206 case FW6_TYPE_WR_RPL:
4207 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4208 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4209 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4211 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4215 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4219 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4227 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4229 struct cpl_abort_req_rss *req = cplhdr(skb);
4231 unsigned int tid = GET_TID(req);
4233 ep = get_ep_from_tid(dev, tid);
4234 /* This EP will be dereferenced in peer_abort() */
4236 printk(KERN_WARNING MOD
4237 "Abort on non-existent endpoint, tid %d\n", tid);
4241 if (is_neg_adv(req->status)) {
4242 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4243 __func__, ep->hwtid, req->status,
4244 neg_adv_str(req->status));
4247 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4250 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4257 * Most upcalls from the T4 Core go to sched() to
4258 * schedule the processing on a work queue.
4260 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4261 [CPL_ACT_ESTABLISH] = sched,
4262 [CPL_ACT_OPEN_RPL] = sched,
4263 [CPL_RX_DATA] = sched,
4264 [CPL_ABORT_RPL_RSS] = sched,
4265 [CPL_ABORT_RPL] = sched,
4266 [CPL_PASS_OPEN_RPL] = sched,
4267 [CPL_CLOSE_LISTSRV_RPL] = sched,
4268 [CPL_PASS_ACCEPT_REQ] = sched,
4269 [CPL_PASS_ESTABLISH] = sched,
4270 [CPL_PEER_CLOSE] = sched,
4271 [CPL_CLOSE_CON_RPL] = sched,
4272 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4273 [CPL_RDMA_TERMINATE] = sched,
4274 [CPL_FW4_ACK] = sched,
4275 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4276 [CPL_FW6_MSG] = fw6_msg,
4277 [CPL_RX_PKT] = sched
4280 int __init c4iw_cm_init(void)
4282 spin_lock_init(&timeout_lock);
4283 skb_queue_head_init(&rxq);
4285 workq = create_singlethread_workqueue("iw_cxgb4");
4292 void c4iw_cm_term(void)
4294 WARN_ON(!list_empty(&timeout_list));
4295 flush_workqueue(workq);
4296 destroy_workqueue(workq);