2 * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
50 #include <rdma/ib_addr.h>
52 #include <libcxgb_cm.h>
56 static char *states[] = {
73 module_param(nocong, int, 0644);
74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
76 static int enable_ecn;
77 module_param(enable_ecn, int, 0644);
78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
80 static int dack_mode = 1;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
84 uint c4iw_max_read_depth = 32;
85 module_param(c4iw_max_read_depth, int, 0644);
86 MODULE_PARM_DESC(c4iw_max_read_depth,
87 "Per-connection max ORD/IRD (default=32)");
89 static int enable_tcp_timestamps;
90 module_param(enable_tcp_timestamps, int, 0644);
91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
93 static int enable_tcp_sack;
94 module_param(enable_tcp_sack, int, 0644);
95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
97 static int enable_tcp_window_scaling = 1;
98 module_param(enable_tcp_window_scaling, int, 0644);
99 MODULE_PARM_DESC(enable_tcp_window_scaling,
100 "Enable tcp window scaling (default=1)");
103 module_param(c4iw_debug, int, 0644);
104 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
106 static int peer2peer = 1;
107 module_param(peer2peer, int, 0644);
108 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
110 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
111 module_param(p2p_type, int, 0644);
112 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
113 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
115 static int ep_timeout_secs = 60;
116 module_param(ep_timeout_secs, int, 0644);
117 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
118 "in seconds (default=60)");
120 static int mpa_rev = 2;
121 module_param(mpa_rev, int, 0644);
122 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
123 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
124 " compliant (default=2)");
126 static int markers_enabled;
127 module_param(markers_enabled, int, 0644);
128 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
130 static int crc_enabled = 1;
131 module_param(crc_enabled, int, 0644);
132 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
134 static int rcv_win = 256 * 1024;
135 module_param(rcv_win, int, 0644);
136 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
138 static int snd_win = 128 * 1024;
139 module_param(snd_win, int, 0644);
140 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
142 static struct workqueue_struct *workq;
144 static struct sk_buff_head rxq;
146 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
147 static void ep_timeout(unsigned long arg);
148 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
149 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
151 static LIST_HEAD(timeout_list);
152 static spinlock_t timeout_lock;
154 static void deref_cm_id(struct c4iw_ep_common *epc)
156 epc->cm_id->rem_ref(epc->cm_id);
158 set_bit(CM_ID_DEREFED, &epc->history);
161 static void ref_cm_id(struct c4iw_ep_common *epc)
163 set_bit(CM_ID_REFED, &epc->history);
164 epc->cm_id->add_ref(epc->cm_id);
167 static void deref_qp(struct c4iw_ep *ep)
169 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
170 clear_bit(QP_REFERENCED, &ep->com.flags);
171 set_bit(QP_DEREFED, &ep->com.history);
174 static void ref_qp(struct c4iw_ep *ep)
176 set_bit(QP_REFERENCED, &ep->com.flags);
177 set_bit(QP_REFED, &ep->com.history);
178 c4iw_qp_add_ref(&ep->com.qp->ibqp);
181 static void start_ep_timer(struct c4iw_ep *ep)
183 PDBG("%s ep %p\n", __func__, ep);
184 if (timer_pending(&ep->timer)) {
185 pr_err("%s timer already started! ep %p\n",
189 clear_bit(TIMEOUT, &ep->com.flags);
190 c4iw_get_ep(&ep->com);
191 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
192 ep->timer.data = (unsigned long)ep;
193 ep->timer.function = ep_timeout;
194 add_timer(&ep->timer);
197 static int stop_ep_timer(struct c4iw_ep *ep)
199 PDBG("%s ep %p stopping\n", __func__, ep);
200 del_timer_sync(&ep->timer);
201 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
202 c4iw_put_ep(&ep->com);
208 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
209 struct l2t_entry *l2e)
213 if (c4iw_fatal_error(rdev)) {
215 PDBG("%s - device in error state - dropping\n", __func__);
218 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
221 else if (error == NET_XMIT_DROP)
223 return error < 0 ? error : 0;
226 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
230 if (c4iw_fatal_error(rdev)) {
232 PDBG("%s - device in error state - dropping\n", __func__);
235 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
238 return error < 0 ? error : 0;
241 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
243 u32 len = roundup(sizeof(struct cpl_tid_release), 16);
245 skb = get_skb(skb, len, GFP_KERNEL);
249 cxgb_mk_tid_release(skb, len, hwtid, 0);
250 c4iw_ofld_send(rdev, skb);
254 static void set_emss(struct c4iw_ep *ep, u16 opt)
256 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
257 ((AF_INET == ep->com.remote_addr.ss_family) ?
258 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
259 sizeof(struct tcphdr);
261 if (TCPOPT_TSTAMP_G(opt))
262 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
266 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
267 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
268 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
272 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
274 enum c4iw_ep_state state;
276 mutex_lock(&epc->mutex);
278 mutex_unlock(&epc->mutex);
282 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
287 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
289 mutex_lock(&epc->mutex);
290 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
291 __state_set(epc, new);
292 mutex_unlock(&epc->mutex);
296 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
302 len = roundup(sizeof(union cpl_wr_size), 16);
303 for (i = 0; i < size; i++) {
304 skb = alloc_skb(len, GFP_KERNEL);
307 skb_queue_tail(ep_skb_list, skb);
311 skb_queue_purge(ep_skb_list);
315 static void *alloc_ep(int size, gfp_t gfp)
317 struct c4iw_ep_common *epc;
319 epc = kzalloc(size, gfp);
321 kref_init(&epc->kref);
322 mutex_init(&epc->mutex);
323 c4iw_init_wr_wait(&epc->wr_wait);
325 PDBG("%s alloc ep %p\n", __func__, epc);
329 static void remove_ep_tid(struct c4iw_ep *ep)
333 spin_lock_irqsave(&ep->com.dev->lock, flags);
334 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
335 if (idr_is_empty(&ep->com.dev->hwtid_idr))
336 wake_up(&ep->com.dev->wait);
337 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
340 static void insert_ep_tid(struct c4iw_ep *ep)
344 spin_lock_irqsave(&ep->com.dev->lock, flags);
345 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
346 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
350 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
352 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
357 spin_lock_irqsave(&dev->lock, flags);
358 ep = idr_find(&dev->hwtid_idr, tid);
360 c4iw_get_ep(&ep->com);
361 spin_unlock_irqrestore(&dev->lock, flags);
366 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
368 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
371 struct c4iw_listen_ep *ep;
374 spin_lock_irqsave(&dev->lock, flags);
375 ep = idr_find(&dev->stid_idr, stid);
377 c4iw_get_ep(&ep->com);
378 spin_unlock_irqrestore(&dev->lock, flags);
382 void _c4iw_free_ep(struct kref *kref)
386 ep = container_of(kref, struct c4iw_ep, com.kref);
387 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
388 if (test_bit(QP_REFERENCED, &ep->com.flags))
390 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
391 if (ep->com.remote_addr.ss_family == AF_INET6) {
392 struct sockaddr_in6 *sin6 =
393 (struct sockaddr_in6 *)
397 ep->com.dev->rdev.lldi.ports[0],
398 (const u32 *)&sin6->sin6_addr.s6_addr,
401 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
402 dst_release(ep->dst);
403 cxgb4_l2t_release(ep->l2t);
405 kfree_skb(ep->mpa_skb);
407 if (!skb_queue_empty(&ep->com.ep_skb_list))
408 skb_queue_purge(&ep->com.ep_skb_list);
412 static void release_ep_resources(struct c4iw_ep *ep)
414 set_bit(RELEASE_RESOURCES, &ep->com.flags);
417 * If we have a hwtid, then remove it from the idr table
418 * so lookups will no longer find this endpoint. Otherwise
419 * we have a race where one thread finds the ep ptr just
420 * before the other thread is freeing the ep memory.
424 c4iw_put_ep(&ep->com);
427 static int status2errno(int status)
432 case CPL_ERR_CONN_RESET:
434 case CPL_ERR_ARP_MISS:
435 return -EHOSTUNREACH;
436 case CPL_ERR_CONN_TIMEDOUT:
438 case CPL_ERR_TCAM_FULL:
440 case CPL_ERR_CONN_EXIST:
448 * Try and reuse skbs already allocated...
450 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
452 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
455 skb_reset_transport_header(skb);
457 skb = alloc_skb(len, gfp);
461 t4_set_arp_err_handler(skb, NULL, NULL);
465 static struct net_device *get_real_dev(struct net_device *egress_dev)
467 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
470 static void arp_failure_discard(void *handle, struct sk_buff *skb)
472 pr_err(MOD "ARP failure\n");
476 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
478 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
483 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
484 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
487 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
491 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
492 release_ep_resources(ep);
496 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
500 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
501 c4iw_put_ep(&ep->parent_ep->com);
502 release_ep_resources(ep);
507 * Fake up a special CPL opcode and call sched() so process_work() will call
508 * _put_ep_safe() in a safe context to free the ep resources. This is needed
509 * because ARP error handlers are called in an ATOMIC context, and
510 * _c4iw_free_ep() needs to block.
512 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
515 struct cpl_act_establish *rpl = cplhdr(skb);
517 /* Set our special ARP_FAILURE opcode */
518 rpl->ot.opcode = cpl;
521 * Save ep in the skb->cb area, after where sched() will save the dev
524 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
525 sched(ep->com.dev, skb);
528 /* Handle an ARP failure for an accept */
529 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
531 struct c4iw_ep *ep = handle;
533 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
536 __state_set(&ep->com, DEAD);
537 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
541 * Handle an ARP failure for an active open.
543 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
545 struct c4iw_ep *ep = handle;
547 printk(KERN_ERR MOD "ARP failure during connect\n");
548 connect_reply_upcall(ep, -EHOSTUNREACH);
549 __state_set(&ep->com, DEAD);
550 if (ep->com.remote_addr.ss_family == AF_INET6) {
551 struct sockaddr_in6 *sin6 =
552 (struct sockaddr_in6 *)&ep->com.local_addr;
553 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
554 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
556 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
557 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
558 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
562 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
565 static void abort_arp_failure(void *handle, struct sk_buff *skb)
568 struct c4iw_ep *ep = handle;
569 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
570 struct cpl_abort_req *req = cplhdr(skb);
572 PDBG("%s rdev %p\n", __func__, rdev);
573 req->cmd = CPL_ABORT_NO_RST;
575 ret = c4iw_ofld_send(rdev, skb);
577 __state_set(&ep->com, DEAD);
578 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
583 static int send_flowc(struct c4iw_ep *ep)
585 struct fw_flowc_wr *flowc;
586 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
588 u16 vlan = ep->l2t->vlan;
594 if (vlan == CPL_L2T_VLAN_NONE)
599 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
601 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
602 FW_FLOWC_WR_NPARAMS_V(nparams));
603 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
604 16)) | FW_WR_FLOWID_V(ep->hwtid));
606 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
607 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
608 (ep->com.dev->rdev.lldi.pf));
609 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
610 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
611 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
612 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
613 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
614 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
615 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
616 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
617 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
618 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
619 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
620 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
621 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
622 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
626 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
627 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
628 flowc->mnemval[8].val = cpu_to_be32(pri);
630 /* Pad WR to 16 byte boundary */
631 flowc->mnemval[8].mnemonic = 0;
632 flowc->mnemval[8].val = 0;
634 for (i = 0; i < 9; i++) {
635 flowc->mnemval[i].r4[0] = 0;
636 flowc->mnemval[i].r4[1] = 0;
637 flowc->mnemval[i].r4[2] = 0;
640 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
641 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
644 static int send_halfclose(struct c4iw_ep *ep)
646 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
647 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
649 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
653 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
654 NULL, arp_failure_discard);
656 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
659 static int send_abort(struct c4iw_ep *ep)
661 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
662 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
664 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
665 if (WARN_ON(!req_skb))
668 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
669 ep, abort_arp_failure);
671 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
674 static int send_connect(struct c4iw_ep *ep)
676 struct cpl_act_open_req *req = NULL;
677 struct cpl_t5_act_open_req *t5req = NULL;
678 struct cpl_t6_act_open_req *t6req = NULL;
679 struct cpl_act_open_req6 *req6 = NULL;
680 struct cpl_t5_act_open_req6 *t5req6 = NULL;
681 struct cpl_t6_act_open_req6 *t6req6 = NULL;
685 unsigned int mtu_idx;
687 int win, sizev4, sizev6, wrlen;
688 struct sockaddr_in *la = (struct sockaddr_in *)
690 struct sockaddr_in *ra = (struct sockaddr_in *)
691 &ep->com.remote_addr;
692 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
694 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
695 &ep->com.remote_addr;
697 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
698 u32 isn = (prandom_u32() & ~7UL) - 1;
700 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
702 sizev4 = sizeof(struct cpl_act_open_req);
703 sizev6 = sizeof(struct cpl_act_open_req6);
706 sizev4 = sizeof(struct cpl_t5_act_open_req);
707 sizev6 = sizeof(struct cpl_t5_act_open_req6);
710 sizev4 = sizeof(struct cpl_t6_act_open_req);
711 sizev6 = sizeof(struct cpl_t6_act_open_req6);
714 pr_err("T%d Chip is not supported\n",
715 CHELSIO_CHIP_VERSION(adapter_type));
719 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
720 roundup(sizev4, 16) :
723 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
725 skb = get_skb(NULL, wrlen, GFP_KERNEL);
727 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
731 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
733 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
734 enable_tcp_timestamps,
735 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
736 wscale = cxgb_compute_wscale(rcv_win);
739 * Specify the largest window that will fit in opt0. The
740 * remainder will be specified in the rx_data_ack.
742 win = ep->rcv_win >> 10;
743 if (win > RCV_BUFSIZ_M)
746 opt0 = (nocong ? NO_CONG_F : 0) |
749 WND_SCALE_V(wscale) |
751 L2T_IDX_V(ep->l2t->idx) |
752 TX_CHAN_V(ep->tx_chan) |
753 SMAC_SEL_V(ep->smac_idx) |
754 DSCP_V(ep->tos >> 2) |
755 ULP_MODE_V(ULP_MODE_TCPDDP) |
757 opt2 = RX_CHANNEL_V(0) |
758 CCTRL_ECN_V(enable_ecn) |
759 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
760 if (enable_tcp_timestamps)
761 opt2 |= TSTAMPS_EN_F;
764 if (wscale && enable_tcp_window_scaling)
765 opt2 |= WND_SCALE_EN_F;
766 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
770 opt2 |= T5_OPT_2_VALID_F;
771 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
775 if (ep->com.remote_addr.ss_family == AF_INET6)
776 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
777 (const u32 *)&la6->sin6_addr.s6_addr, 1);
779 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
781 if (ep->com.remote_addr.ss_family == AF_INET) {
782 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
784 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
788 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
790 INIT_TP_WR(t5req, 0);
791 req = (struct cpl_act_open_req *)t5req;
794 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
796 INIT_TP_WR(t6req, 0);
797 req = (struct cpl_act_open_req *)t6req;
798 t5req = (struct cpl_t5_act_open_req *)t6req;
801 pr_err("T%d Chip is not supported\n",
802 CHELSIO_CHIP_VERSION(adapter_type));
807 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
808 ((ep->rss_qid<<14) | ep->atid)));
809 req->local_port = la->sin_port;
810 req->peer_port = ra->sin_port;
811 req->local_ip = la->sin_addr.s_addr;
812 req->peer_ip = ra->sin_addr.s_addr;
813 req->opt0 = cpu_to_be64(opt0);
815 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
816 req->params = cpu_to_be32(cxgb4_select_ntuple(
817 ep->com.dev->rdev.lldi.ports[0],
819 req->opt2 = cpu_to_be32(opt2);
821 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
823 ep->com.dev->rdev.lldi.ports[0],
825 t5req->rsvd = cpu_to_be32(isn);
826 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
827 t5req->opt2 = cpu_to_be32(opt2);
830 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
832 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
836 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
838 INIT_TP_WR(t5req6, 0);
839 req6 = (struct cpl_act_open_req6 *)t5req6;
842 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
844 INIT_TP_WR(t6req6, 0);
845 req6 = (struct cpl_act_open_req6 *)t6req6;
846 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
849 pr_err("T%d Chip is not supported\n",
850 CHELSIO_CHIP_VERSION(adapter_type));
855 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
856 ((ep->rss_qid<<14)|ep->atid)));
857 req6->local_port = la6->sin6_port;
858 req6->peer_port = ra6->sin6_port;
859 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
860 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
861 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
862 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
863 req6->opt0 = cpu_to_be64(opt0);
865 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
866 req6->params = cpu_to_be32(cxgb4_select_ntuple(
867 ep->com.dev->rdev.lldi.ports[0],
869 req6->opt2 = cpu_to_be32(opt2);
871 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
873 ep->com.dev->rdev.lldi.ports[0],
875 t5req6->rsvd = cpu_to_be32(isn);
876 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
877 t5req6->opt2 = cpu_to_be32(opt2);
881 set_bit(ACT_OPEN_REQ, &ep->com.history);
882 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
884 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
885 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
886 (const u32 *)&la6->sin6_addr.s6_addr, 1);
890 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
893 int mpalen, wrlen, ret;
894 struct fw_ofld_tx_data_wr *req;
895 struct mpa_message *mpa;
896 struct mpa_v2_conn_params mpa_v2_params;
898 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
900 BUG_ON(skb_cloned(skb));
902 mpalen = sizeof(*mpa) + ep->plen;
903 if (mpa_rev_to_use == 2)
904 mpalen += sizeof(struct mpa_v2_conn_params);
905 wrlen = roundup(mpalen + sizeof *req, 16);
906 skb = get_skb(skb, wrlen, GFP_KERNEL);
908 connect_reply_upcall(ep, -ENOMEM);
911 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
913 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
914 memset(req, 0, wrlen);
915 req->op_to_immdlen = cpu_to_be32(
916 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
918 FW_WR_IMMDLEN_V(mpalen));
919 req->flowid_len16 = cpu_to_be32(
920 FW_WR_FLOWID_V(ep->hwtid) |
921 FW_WR_LEN16_V(wrlen >> 4));
922 req->plen = cpu_to_be32(mpalen);
923 req->tunnel_to_proxy = cpu_to_be32(
924 FW_OFLD_TX_DATA_WR_FLUSH_F |
925 FW_OFLD_TX_DATA_WR_SHOVE_F);
927 mpa = (struct mpa_message *)(req + 1);
928 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
932 mpa->flags |= MPA_CRC;
933 if (markers_enabled) {
934 mpa->flags |= MPA_MARKERS;
935 ep->mpa_attr.recv_marker_enabled = 1;
937 ep->mpa_attr.recv_marker_enabled = 0;
939 if (mpa_rev_to_use == 2)
940 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
942 mpa->private_data_size = htons(ep->plen);
943 mpa->revision = mpa_rev_to_use;
944 if (mpa_rev_to_use == 1) {
945 ep->tried_with_mpa_v1 = 1;
946 ep->retry_with_mpa_v1 = 0;
949 if (mpa_rev_to_use == 2) {
950 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
951 sizeof (struct mpa_v2_conn_params));
952 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
954 mpa_v2_params.ird = htons((u16)ep->ird);
955 mpa_v2_params.ord = htons((u16)ep->ord);
958 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
959 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
961 htons(MPA_V2_RDMA_WRITE_RTR);
962 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
964 htons(MPA_V2_RDMA_READ_RTR);
966 memcpy(mpa->private_data, &mpa_v2_params,
967 sizeof(struct mpa_v2_conn_params));
970 memcpy(mpa->private_data +
971 sizeof(struct mpa_v2_conn_params),
972 ep->mpa_pkt + sizeof(*mpa), ep->plen);
975 memcpy(mpa->private_data,
976 ep->mpa_pkt + sizeof(*mpa), ep->plen);
979 * Reference the mpa skb. This ensures the data area
980 * will remain in memory until the hw acks the tx.
981 * Function fw4_ack() will deref it.
984 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
987 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
991 __state_set(&ep->com, MPA_REQ_SENT);
992 ep->mpa_attr.initiator = 1;
993 ep->snd_seq += mpalen;
997 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1000 struct fw_ofld_tx_data_wr *req;
1001 struct mpa_message *mpa;
1002 struct sk_buff *skb;
1003 struct mpa_v2_conn_params mpa_v2_params;
1005 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1007 mpalen = sizeof(*mpa) + plen;
1008 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1009 mpalen += sizeof(struct mpa_v2_conn_params);
1010 wrlen = roundup(mpalen + sizeof *req, 16);
1012 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1014 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1017 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1019 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1020 memset(req, 0, wrlen);
1021 req->op_to_immdlen = cpu_to_be32(
1022 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1024 FW_WR_IMMDLEN_V(mpalen));
1025 req->flowid_len16 = cpu_to_be32(
1026 FW_WR_FLOWID_V(ep->hwtid) |
1027 FW_WR_LEN16_V(wrlen >> 4));
1028 req->plen = cpu_to_be32(mpalen);
1029 req->tunnel_to_proxy = cpu_to_be32(
1030 FW_OFLD_TX_DATA_WR_FLUSH_F |
1031 FW_OFLD_TX_DATA_WR_SHOVE_F);
1033 mpa = (struct mpa_message *)(req + 1);
1034 memset(mpa, 0, sizeof(*mpa));
1035 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1036 mpa->flags = MPA_REJECT;
1037 mpa->revision = ep->mpa_attr.version;
1038 mpa->private_data_size = htons(plen);
1040 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1041 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1042 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1043 sizeof (struct mpa_v2_conn_params));
1044 mpa_v2_params.ird = htons(((u16)ep->ird) |
1045 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1047 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1049 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1050 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1051 FW_RI_INIT_P2PTYPE_READ_REQ ?
1052 MPA_V2_RDMA_READ_RTR : 0) : 0));
1053 memcpy(mpa->private_data, &mpa_v2_params,
1054 sizeof(struct mpa_v2_conn_params));
1057 memcpy(mpa->private_data +
1058 sizeof(struct mpa_v2_conn_params), pdata, plen);
1061 memcpy(mpa->private_data, pdata, plen);
1064 * Reference the mpa skb again. This ensures the data area
1065 * will remain in memory until the hw acks the tx.
1066 * Function fw4_ack() will deref it.
1069 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1070 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1071 BUG_ON(ep->mpa_skb);
1073 ep->snd_seq += mpalen;
1074 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1077 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1080 struct fw_ofld_tx_data_wr *req;
1081 struct mpa_message *mpa;
1082 struct sk_buff *skb;
1083 struct mpa_v2_conn_params mpa_v2_params;
1085 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1087 mpalen = sizeof(*mpa) + plen;
1088 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1089 mpalen += sizeof(struct mpa_v2_conn_params);
1090 wrlen = roundup(mpalen + sizeof *req, 16);
1092 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1094 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1097 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1099 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1100 memset(req, 0, wrlen);
1101 req->op_to_immdlen = cpu_to_be32(
1102 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1104 FW_WR_IMMDLEN_V(mpalen));
1105 req->flowid_len16 = cpu_to_be32(
1106 FW_WR_FLOWID_V(ep->hwtid) |
1107 FW_WR_LEN16_V(wrlen >> 4));
1108 req->plen = cpu_to_be32(mpalen);
1109 req->tunnel_to_proxy = cpu_to_be32(
1110 FW_OFLD_TX_DATA_WR_FLUSH_F |
1111 FW_OFLD_TX_DATA_WR_SHOVE_F);
1113 mpa = (struct mpa_message *)(req + 1);
1114 memset(mpa, 0, sizeof(*mpa));
1115 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1117 if (ep->mpa_attr.crc_enabled)
1118 mpa->flags |= MPA_CRC;
1119 if (ep->mpa_attr.recv_marker_enabled)
1120 mpa->flags |= MPA_MARKERS;
1121 mpa->revision = ep->mpa_attr.version;
1122 mpa->private_data_size = htons(plen);
1124 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1125 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1126 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1127 sizeof (struct mpa_v2_conn_params));
1128 mpa_v2_params.ird = htons((u16)ep->ird);
1129 mpa_v2_params.ord = htons((u16)ep->ord);
1130 if (peer2peer && (ep->mpa_attr.p2p_type !=
1131 FW_RI_INIT_P2PTYPE_DISABLED)) {
1132 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1134 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1135 mpa_v2_params.ord |=
1136 htons(MPA_V2_RDMA_WRITE_RTR);
1137 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1138 mpa_v2_params.ord |=
1139 htons(MPA_V2_RDMA_READ_RTR);
1142 memcpy(mpa->private_data, &mpa_v2_params,
1143 sizeof(struct mpa_v2_conn_params));
1146 memcpy(mpa->private_data +
1147 sizeof(struct mpa_v2_conn_params), pdata, plen);
1150 memcpy(mpa->private_data, pdata, plen);
1153 * Reference the mpa skb. This ensures the data area
1154 * will remain in memory until the hw acks the tx.
1155 * Function fw4_ack() will deref it.
1158 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1160 __state_set(&ep->com, MPA_REP_SENT);
1161 ep->snd_seq += mpalen;
1162 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1165 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1168 struct cpl_act_establish *req = cplhdr(skb);
1169 unsigned int tid = GET_TID(req);
1170 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1171 struct tid_info *t = dev->rdev.lldi.tids;
1174 ep = lookup_atid(t, atid);
1176 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1177 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1179 mutex_lock(&ep->com.mutex);
1180 dst_confirm(ep->dst);
1182 /* setup the hwtid for this connection */
1184 cxgb4_insert_tid(t, ep, tid);
1187 ep->snd_seq = be32_to_cpu(req->snd_isn);
1188 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1190 set_emss(ep, ntohs(req->tcp_opt));
1192 /* dealloc the atid */
1193 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1194 cxgb4_free_atid(t, atid);
1195 set_bit(ACT_ESTAB, &ep->com.history);
1197 /* start MPA negotiation */
1198 ret = send_flowc(ep);
1201 if (ep->retry_with_mpa_v1)
1202 ret = send_mpa_req(ep, skb, 1);
1204 ret = send_mpa_req(ep, skb, mpa_rev);
1207 mutex_unlock(&ep->com.mutex);
1210 mutex_unlock(&ep->com.mutex);
1211 connect_reply_upcall(ep, -ENOMEM);
1212 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1216 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1218 struct iw_cm_event event;
1220 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1221 memset(&event, 0, sizeof(event));
1222 event.event = IW_CM_EVENT_CLOSE;
1223 event.status = status;
1224 if (ep->com.cm_id) {
1225 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1226 ep, ep->com.cm_id, ep->hwtid);
1227 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1228 deref_cm_id(&ep->com);
1229 set_bit(CLOSE_UPCALL, &ep->com.history);
1233 static void peer_close_upcall(struct c4iw_ep *ep)
1235 struct iw_cm_event event;
1237 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1238 memset(&event, 0, sizeof(event));
1239 event.event = IW_CM_EVENT_DISCONNECT;
1240 if (ep->com.cm_id) {
1241 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1242 ep, ep->com.cm_id, ep->hwtid);
1243 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1244 set_bit(DISCONN_UPCALL, &ep->com.history);
1248 static void peer_abort_upcall(struct c4iw_ep *ep)
1250 struct iw_cm_event event;
1252 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1253 memset(&event, 0, sizeof(event));
1254 event.event = IW_CM_EVENT_CLOSE;
1255 event.status = -ECONNRESET;
1256 if (ep->com.cm_id) {
1257 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1258 ep->com.cm_id, ep->hwtid);
1259 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1260 deref_cm_id(&ep->com);
1261 set_bit(ABORT_UPCALL, &ep->com.history);
1265 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1267 struct iw_cm_event event;
1269 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1270 memset(&event, 0, sizeof(event));
1271 event.event = IW_CM_EVENT_CONNECT_REPLY;
1272 event.status = status;
1273 memcpy(&event.local_addr, &ep->com.local_addr,
1274 sizeof(ep->com.local_addr));
1275 memcpy(&event.remote_addr, &ep->com.remote_addr,
1276 sizeof(ep->com.remote_addr));
1278 if ((status == 0) || (status == -ECONNREFUSED)) {
1279 if (!ep->tried_with_mpa_v1) {
1280 /* this means MPA_v2 is used */
1281 event.ord = ep->ird;
1282 event.ird = ep->ord;
1283 event.private_data_len = ep->plen -
1284 sizeof(struct mpa_v2_conn_params);
1285 event.private_data = ep->mpa_pkt +
1286 sizeof(struct mpa_message) +
1287 sizeof(struct mpa_v2_conn_params);
1289 /* this means MPA_v1 is used */
1290 event.ord = cur_max_read_depth(ep->com.dev);
1291 event.ird = cur_max_read_depth(ep->com.dev);
1292 event.private_data_len = ep->plen;
1293 event.private_data = ep->mpa_pkt +
1294 sizeof(struct mpa_message);
1298 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1300 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1301 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1304 deref_cm_id(&ep->com);
1307 static int connect_request_upcall(struct c4iw_ep *ep)
1309 struct iw_cm_event event;
1312 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1313 memset(&event, 0, sizeof(event));
1314 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1315 memcpy(&event.local_addr, &ep->com.local_addr,
1316 sizeof(ep->com.local_addr));
1317 memcpy(&event.remote_addr, &ep->com.remote_addr,
1318 sizeof(ep->com.remote_addr));
1319 event.provider_data = ep;
1320 if (!ep->tried_with_mpa_v1) {
1321 /* this means MPA_v2 is used */
1322 event.ord = ep->ord;
1323 event.ird = ep->ird;
1324 event.private_data_len = ep->plen -
1325 sizeof(struct mpa_v2_conn_params);
1326 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1327 sizeof(struct mpa_v2_conn_params);
1329 /* this means MPA_v1 is used. Send max supported */
1330 event.ord = cur_max_read_depth(ep->com.dev);
1331 event.ird = cur_max_read_depth(ep->com.dev);
1332 event.private_data_len = ep->plen;
1333 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1335 c4iw_get_ep(&ep->com);
1336 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1339 c4iw_put_ep(&ep->com);
1340 set_bit(CONNREQ_UPCALL, &ep->com.history);
1341 c4iw_put_ep(&ep->parent_ep->com);
1345 static void established_upcall(struct c4iw_ep *ep)
1347 struct iw_cm_event event;
1349 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1350 memset(&event, 0, sizeof(event));
1351 event.event = IW_CM_EVENT_ESTABLISHED;
1352 event.ird = ep->ord;
1353 event.ord = ep->ird;
1354 if (ep->com.cm_id) {
1355 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1356 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1357 set_bit(ESTAB_UPCALL, &ep->com.history);
1361 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1363 struct sk_buff *skb;
1364 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1367 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1368 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1370 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1375 * If we couldn't specify the entire rcv window at connection setup
1376 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1377 * then add the overage in to the credits returned.
1379 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1380 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1382 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1383 RX_DACK_MODE_V(dack_mode);
1385 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1388 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1392 #define RELAXED_IRD_NEGOTIATION 1
1395 * process_mpa_reply - process streaming mode MPA reply
1399 * 0 upon success indicating a connect request was delivered to the ULP
1400 * or the mpa request is incomplete but valid so far.
1402 * 1 if a failure requires the caller to close the connection.
1404 * 2 if a failure requires the caller to abort the connection.
1406 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1408 struct mpa_message *mpa;
1409 struct mpa_v2_conn_params *mpa_v2_params;
1411 u16 resp_ird, resp_ord;
1412 u8 rtr_mismatch = 0, insuff_ird = 0;
1413 struct c4iw_qp_attributes attrs;
1414 enum c4iw_qp_attr_mask mask;
1418 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1421 * If we get more than the supported amount of private data
1422 * then we must fail this connection.
1424 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1426 goto err_stop_timer;
1430 * copy the new data into our accumulation buffer.
1432 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1434 ep->mpa_pkt_len += skb->len;
1437 * if we don't even have the mpa message, then bail.
1439 if (ep->mpa_pkt_len < sizeof(*mpa))
1441 mpa = (struct mpa_message *) ep->mpa_pkt;
1443 /* Validate MPA header. */
1444 if (mpa->revision > mpa_rev) {
1445 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1446 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1448 goto err_stop_timer;
1450 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1452 goto err_stop_timer;
1455 plen = ntohs(mpa->private_data_size);
1458 * Fail if there's too much private data.
1460 if (plen > MPA_MAX_PRIVATE_DATA) {
1462 goto err_stop_timer;
1466 * If plen does not account for pkt size
1468 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1470 goto err_stop_timer;
1473 ep->plen = (u8) plen;
1476 * If we don't have all the pdata yet, then bail.
1477 * We'll continue process when more data arrives.
1479 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1482 if (mpa->flags & MPA_REJECT) {
1483 err = -ECONNREFUSED;
1484 goto err_stop_timer;
1488 * Stop mpa timer. If it expired, then
1489 * we ignore the MPA reply. process_timeout()
1490 * will abort the connection.
1492 if (stop_ep_timer(ep))
1496 * If we get here we have accumulated the entire mpa
1497 * start reply message including private data. And
1498 * the MPA header is valid.
1500 __state_set(&ep->com, FPDU_MODE);
1501 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1502 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1503 ep->mpa_attr.version = mpa->revision;
1504 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1506 if (mpa->revision == 2) {
1507 ep->mpa_attr.enhanced_rdma_conn =
1508 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1509 if (ep->mpa_attr.enhanced_rdma_conn) {
1510 mpa_v2_params = (struct mpa_v2_conn_params *)
1511 (ep->mpa_pkt + sizeof(*mpa));
1512 resp_ird = ntohs(mpa_v2_params->ird) &
1513 MPA_V2_IRD_ORD_MASK;
1514 resp_ord = ntohs(mpa_v2_params->ord) &
1515 MPA_V2_IRD_ORD_MASK;
1516 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1517 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1520 * This is a double-check. Ideally, below checks are
1521 * not required since ird/ord stuff has been taken
1522 * care of in c4iw_accept_cr
1524 if (ep->ird < resp_ord) {
1525 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1526 ep->com.dev->rdev.lldi.max_ordird_qp)
1530 } else if (ep->ird > resp_ord) {
1533 if (ep->ord > resp_ird) {
1534 if (RELAXED_IRD_NEGOTIATION)
1545 if (ntohs(mpa_v2_params->ird) &
1546 MPA_V2_PEER2PEER_MODEL) {
1547 if (ntohs(mpa_v2_params->ord) &
1548 MPA_V2_RDMA_WRITE_RTR)
1549 ep->mpa_attr.p2p_type =
1550 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1551 else if (ntohs(mpa_v2_params->ord) &
1552 MPA_V2_RDMA_READ_RTR)
1553 ep->mpa_attr.p2p_type =
1554 FW_RI_INIT_P2PTYPE_READ_REQ;
1557 } else if (mpa->revision == 1)
1559 ep->mpa_attr.p2p_type = p2p_type;
1561 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1562 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1563 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1564 ep->mpa_attr.recv_marker_enabled,
1565 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1566 ep->mpa_attr.p2p_type, p2p_type);
1569 * If responder's RTR does not match with that of initiator, assign
1570 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1571 * generated when moving QP to RTS state.
1572 * A TERM message will be sent after QP has moved to RTS state
1574 if ((ep->mpa_attr.version == 2) && peer2peer &&
1575 (ep->mpa_attr.p2p_type != p2p_type)) {
1576 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1580 attrs.mpa_attr = ep->mpa_attr;
1581 attrs.max_ird = ep->ird;
1582 attrs.max_ord = ep->ord;
1583 attrs.llp_stream_handle = ep;
1584 attrs.next_state = C4IW_QP_STATE_RTS;
1586 mask = C4IW_QP_ATTR_NEXT_STATE |
1587 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1588 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1590 /* bind QP and TID with INIT_WR */
1591 err = c4iw_modify_qp(ep->com.qp->rhp,
1592 ep->com.qp, mask, &attrs, 1);
1597 * If responder's RTR requirement did not match with what initiator
1598 * supports, generate TERM message
1601 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1602 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1603 attrs.ecode = MPA_NOMATCH_RTR;
1604 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1605 attrs.send_term = 1;
1606 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1607 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1614 * Generate TERM if initiator IRD is not sufficient for responder
1615 * provided ORD. Currently, we do the same behaviour even when
1616 * responder provided IRD is also not sufficient as regards to
1620 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1622 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1623 attrs.ecode = MPA_INSUFF_IRD;
1624 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1625 attrs.send_term = 1;
1626 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1627 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1638 connect_reply_upcall(ep, err);
1643 * process_mpa_request - process streaming mode MPA request
1647 * 0 upon success indicating a connect request was delivered to the ULP
1648 * or the mpa request is incomplete but valid so far.
1650 * 1 if a failure requires the caller to close the connection.
1652 * 2 if a failure requires the caller to abort the connection.
1654 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1656 struct mpa_message *mpa;
1657 struct mpa_v2_conn_params *mpa_v2_params;
1660 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1663 * If we get more than the supported amount of private data
1664 * then we must fail this connection.
1666 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1667 goto err_stop_timer;
1669 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1672 * Copy the new data into our accumulation buffer.
1674 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1676 ep->mpa_pkt_len += skb->len;
1679 * If we don't even have the mpa message, then bail.
1680 * We'll continue process when more data arrives.
1682 if (ep->mpa_pkt_len < sizeof(*mpa))
1685 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1686 mpa = (struct mpa_message *) ep->mpa_pkt;
1689 * Validate MPA Header.
1691 if (mpa->revision > mpa_rev) {
1692 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1693 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1694 goto err_stop_timer;
1697 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1698 goto err_stop_timer;
1700 plen = ntohs(mpa->private_data_size);
1703 * Fail if there's too much private data.
1705 if (plen > MPA_MAX_PRIVATE_DATA)
1706 goto err_stop_timer;
1709 * If plen does not account for pkt size
1711 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1712 goto err_stop_timer;
1713 ep->plen = (u8) plen;
1716 * If we don't have all the pdata yet, then bail.
1718 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1722 * If we get here we have accumulated the entire mpa
1723 * start reply message including private data.
1725 ep->mpa_attr.initiator = 0;
1726 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1727 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1728 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1729 ep->mpa_attr.version = mpa->revision;
1730 if (mpa->revision == 1)
1731 ep->tried_with_mpa_v1 = 1;
1732 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1734 if (mpa->revision == 2) {
1735 ep->mpa_attr.enhanced_rdma_conn =
1736 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1737 if (ep->mpa_attr.enhanced_rdma_conn) {
1738 mpa_v2_params = (struct mpa_v2_conn_params *)
1739 (ep->mpa_pkt + sizeof(*mpa));
1740 ep->ird = ntohs(mpa_v2_params->ird) &
1741 MPA_V2_IRD_ORD_MASK;
1742 ep->ird = min_t(u32, ep->ird,
1743 cur_max_read_depth(ep->com.dev));
1744 ep->ord = ntohs(mpa_v2_params->ord) &
1745 MPA_V2_IRD_ORD_MASK;
1746 ep->ord = min_t(u32, ep->ord,
1747 cur_max_read_depth(ep->com.dev));
1748 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1750 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1752 if (ntohs(mpa_v2_params->ord) &
1753 MPA_V2_RDMA_WRITE_RTR)
1754 ep->mpa_attr.p2p_type =
1755 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1756 else if (ntohs(mpa_v2_params->ord) &
1757 MPA_V2_RDMA_READ_RTR)
1758 ep->mpa_attr.p2p_type =
1759 FW_RI_INIT_P2PTYPE_READ_REQ;
1762 } else if (mpa->revision == 1)
1764 ep->mpa_attr.p2p_type = p2p_type;
1766 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1767 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1768 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1769 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1770 ep->mpa_attr.p2p_type);
1772 __state_set(&ep->com, MPA_REQ_RCVD);
1775 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1776 if (ep->parent_ep->com.state != DEAD) {
1777 if (connect_request_upcall(ep))
1778 goto err_unlock_parent;
1780 goto err_unlock_parent;
1782 mutex_unlock(&ep->parent_ep->com.mutex);
1786 mutex_unlock(&ep->parent_ep->com.mutex);
1789 (void)stop_ep_timer(ep);
1794 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1797 struct cpl_rx_data *hdr = cplhdr(skb);
1798 unsigned int dlen = ntohs(hdr->len);
1799 unsigned int tid = GET_TID(hdr);
1800 __u8 status = hdr->status;
1803 ep = get_ep_from_tid(dev, tid);
1806 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1807 skb_pull(skb, sizeof(*hdr));
1808 skb_trim(skb, dlen);
1809 mutex_lock(&ep->com.mutex);
1811 switch (ep->com.state) {
1813 update_rx_credits(ep, dlen);
1814 ep->rcv_seq += dlen;
1815 disconnect = process_mpa_reply(ep, skb);
1818 update_rx_credits(ep, dlen);
1819 ep->rcv_seq += dlen;
1820 disconnect = process_mpa_request(ep, skb);
1823 struct c4iw_qp_attributes attrs;
1825 update_rx_credits(ep, dlen);
1826 BUG_ON(!ep->com.qp);
1828 pr_err("%s Unexpected streaming data." \
1829 " qpid %u ep %p state %d tid %u status %d\n",
1830 __func__, ep->com.qp->wq.sq.qid, ep,
1831 ep->com.state, ep->hwtid, status);
1832 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1833 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1834 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1841 mutex_unlock(&ep->com.mutex);
1843 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1844 c4iw_put_ep(&ep->com);
1848 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1851 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1853 unsigned int tid = GET_TID(rpl);
1855 ep = get_ep_from_tid(dev, tid);
1857 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1860 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1861 mutex_lock(&ep->com.mutex);
1862 switch (ep->com.state) {
1864 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1865 __state_set(&ep->com, DEAD);
1869 printk(KERN_ERR "%s ep %p state %d\n",
1870 __func__, ep, ep->com.state);
1873 mutex_unlock(&ep->com.mutex);
1876 close_complete_upcall(ep, -ECONNRESET);
1877 release_ep_resources(ep);
1879 c4iw_put_ep(&ep->com);
1883 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1885 struct sk_buff *skb;
1886 struct fw_ofld_connection_wr *req;
1887 unsigned int mtu_idx;
1889 struct sockaddr_in *sin;
1892 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1893 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1894 memset(req, 0, sizeof(*req));
1895 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1896 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1897 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1898 ep->com.dev->rdev.lldi.ports[0],
1900 sin = (struct sockaddr_in *)&ep->com.local_addr;
1901 req->le.lport = sin->sin_port;
1902 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1903 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1904 req->le.pport = sin->sin_port;
1905 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1906 req->tcb.t_state_to_astid =
1907 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1908 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1909 req->tcb.cplrxdataack_cplpassacceptrpl =
1910 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1911 req->tcb.tx_max = (__force __be32) jiffies;
1912 req->tcb.rcv_adv = htons(1);
1913 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1914 enable_tcp_timestamps,
1915 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1916 wscale = cxgb_compute_wscale(rcv_win);
1919 * Specify the largest window that will fit in opt0. The
1920 * remainder will be specified in the rx_data_ack.
1922 win = ep->rcv_win >> 10;
1923 if (win > RCV_BUFSIZ_M)
1926 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1927 (nocong ? NO_CONG_F : 0) |
1930 WND_SCALE_V(wscale) |
1931 MSS_IDX_V(mtu_idx) |
1932 L2T_IDX_V(ep->l2t->idx) |
1933 TX_CHAN_V(ep->tx_chan) |
1934 SMAC_SEL_V(ep->smac_idx) |
1935 DSCP_V(ep->tos >> 2) |
1936 ULP_MODE_V(ULP_MODE_TCPDDP) |
1938 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1939 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1941 CCTRL_ECN_V(enable_ecn) |
1942 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1943 if (enable_tcp_timestamps)
1944 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1945 if (enable_tcp_sack)
1946 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1947 if (wscale && enable_tcp_window_scaling)
1948 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1949 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1950 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1951 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1952 set_bit(ACT_OFLD_CONN, &ep->com.history);
1953 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1957 * Some of the error codes above implicitly indicate that there is no TID
1958 * allocated with the result of an ACT_OPEN. We use this predicate to make
1961 static inline int act_open_has_tid(int status)
1963 return (status != CPL_ERR_TCAM_PARITY &&
1964 status != CPL_ERR_TCAM_MISS &&
1965 status != CPL_ERR_TCAM_FULL &&
1966 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1967 status != CPL_ERR_CONN_EXIST);
1970 static char *neg_adv_str(unsigned int status)
1973 case CPL_ERR_RTX_NEG_ADVICE:
1974 return "Retransmit timeout";
1975 case CPL_ERR_PERSIST_NEG_ADVICE:
1976 return "Persist timeout";
1977 case CPL_ERR_KEEPALV_NEG_ADVICE:
1978 return "Keepalive timeout";
1984 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1986 ep->snd_win = snd_win;
1987 ep->rcv_win = rcv_win;
1988 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
1991 #define ACT_OPEN_RETRY_COUNT 2
1993 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1994 struct dst_entry *dst, struct c4iw_dev *cdev,
1995 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
1997 struct neighbour *n;
1999 struct net_device *pdev;
2001 n = dst_neigh_lookup(dst, peer_ip);
2007 if (n->dev->flags & IFF_LOOPBACK) {
2009 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2010 else if (IS_ENABLED(CONFIG_IPV6))
2011 for_each_netdev(&init_net, pdev) {
2012 if (ipv6_chk_addr(&init_net,
2013 (struct in6_addr *)peer_ip,
2024 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2025 n, pdev, rt_tos2priority(tos));
2030 ep->mtu = pdev->mtu;
2031 ep->tx_chan = cxgb4_port_chan(pdev);
2032 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2033 cxgb4_port_viid(pdev));
2034 step = cdev->rdev.lldi.ntxq /
2035 cdev->rdev.lldi.nchan;
2036 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2037 step = cdev->rdev.lldi.nrxq /
2038 cdev->rdev.lldi.nchan;
2039 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2040 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2041 cxgb4_port_idx(pdev) * step];
2042 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2045 pdev = get_real_dev(n->dev);
2046 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2047 n, pdev, rt_tos2priority(tos));
2050 ep->mtu = dst_mtu(dst);
2051 ep->tx_chan = cxgb4_port_chan(pdev);
2052 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2053 cxgb4_port_viid(pdev));
2054 step = cdev->rdev.lldi.ntxq /
2055 cdev->rdev.lldi.nchan;
2056 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2057 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2058 step = cdev->rdev.lldi.nrxq /
2059 cdev->rdev.lldi.nchan;
2060 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2061 cxgb4_port_idx(pdev) * step];
2062 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2065 ep->retry_with_mpa_v1 = 0;
2066 ep->tried_with_mpa_v1 = 0;
2078 static int c4iw_reconnect(struct c4iw_ep *ep)
2082 struct sockaddr_in *laddr = (struct sockaddr_in *)
2083 &ep->com.cm_id->m_local_addr;
2084 struct sockaddr_in *raddr = (struct sockaddr_in *)
2085 &ep->com.cm_id->m_remote_addr;
2086 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2087 &ep->com.cm_id->m_local_addr;
2088 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2089 &ep->com.cm_id->m_remote_addr;
2093 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2094 init_timer(&ep->timer);
2095 c4iw_init_wr_wait(&ep->com.wr_wait);
2097 /* When MPA revision is different on nodes, the node with MPA_rev=2
2098 * tries to reconnect with MPA_rev 1 for the same EP through
2099 * c4iw_reconnect(), where the same EP is assigned with new tid for
2100 * further connection establishment. As we are using the same EP pointer
2101 * for reconnect, few skbs are used during the previous c4iw_connect(),
2102 * which leaves the EP with inadequate skbs for further
2103 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2104 * skb_list() during peer_abort(). Allocate skbs which is already used.
2106 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2107 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2113 * Allocate an active TID to initiate a TCP connection.
2115 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2116 if (ep->atid == -1) {
2117 pr_err("%s - cannot alloc atid.\n", __func__);
2121 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2124 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2125 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2126 laddr->sin_addr.s_addr,
2127 raddr->sin_addr.s_addr,
2129 raddr->sin_port, ep->com.cm_id->tos);
2131 ra = (__u8 *)&raddr->sin_addr;
2133 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2135 laddr6->sin6_addr.s6_addr,
2136 raddr6->sin6_addr.s6_addr,
2140 raddr6->sin6_scope_id);
2142 ra = (__u8 *)&raddr6->sin6_addr;
2145 pr_err("%s - cannot find route.\n", __func__);
2146 err = -EHOSTUNREACH;
2149 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2150 ep->com.dev->rdev.lldi.adapter_type,
2151 ep->com.cm_id->tos);
2153 pr_err("%s - cannot alloc l2e.\n", __func__);
2157 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2158 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2161 state_set(&ep->com, CONNECTING);
2162 ep->tos = ep->com.cm_id->tos;
2164 /* send connect request to rnic */
2165 err = send_connect(ep);
2169 cxgb4_l2t_release(ep->l2t);
2171 dst_release(ep->dst);
2173 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2174 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2177 * remember to send notification to upper layer.
2178 * We are in here so the upper layer is not aware that this is
2179 * re-connect attempt and so, upper layer is still waiting for
2180 * response of 1st connect request.
2182 connect_reply_upcall(ep, -ECONNRESET);
2184 c4iw_put_ep(&ep->com);
2189 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2192 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2193 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2194 ntohl(rpl->atid_status)));
2195 struct tid_info *t = dev->rdev.lldi.tids;
2196 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2197 struct sockaddr_in *la;
2198 struct sockaddr_in *ra;
2199 struct sockaddr_in6 *la6;
2200 struct sockaddr_in6 *ra6;
2203 ep = lookup_atid(t, atid);
2204 la = (struct sockaddr_in *)&ep->com.local_addr;
2205 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2206 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2207 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2209 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2210 status, status2errno(status));
2212 if (cxgb_is_neg_adv(status)) {
2213 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2214 __func__, atid, status, neg_adv_str(status));
2215 ep->stats.connect_neg_adv++;
2216 mutex_lock(&dev->rdev.stats.lock);
2217 dev->rdev.stats.neg_adv++;
2218 mutex_unlock(&dev->rdev.stats.lock);
2222 set_bit(ACT_OPEN_RPL, &ep->com.history);
2225 * Log interesting failures.
2228 case CPL_ERR_CONN_RESET:
2229 case CPL_ERR_CONN_TIMEDOUT:
2231 case CPL_ERR_TCAM_FULL:
2232 mutex_lock(&dev->rdev.stats.lock);
2233 dev->rdev.stats.tcam_full++;
2234 mutex_unlock(&dev->rdev.stats.lock);
2235 if (ep->com.local_addr.ss_family == AF_INET &&
2236 dev->rdev.lldi.enable_fw_ofld_conn) {
2237 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2238 ntohl(rpl->atid_status))));
2244 case CPL_ERR_CONN_EXIST:
2245 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2246 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2247 if (ep->com.remote_addr.ss_family == AF_INET6) {
2248 struct sockaddr_in6 *sin6 =
2249 (struct sockaddr_in6 *)
2250 &ep->com.local_addr;
2252 ep->com.dev->rdev.lldi.ports[0],
2254 &sin6->sin6_addr.s6_addr, 1);
2256 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2258 cxgb4_free_atid(t, atid);
2259 dst_release(ep->dst);
2260 cxgb4_l2t_release(ep->l2t);
2266 if (ep->com.local_addr.ss_family == AF_INET) {
2267 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2268 atid, status, status2errno(status),
2269 &la->sin_addr.s_addr, ntohs(la->sin_port),
2270 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2272 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2273 atid, status, status2errno(status),
2274 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2275 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2281 connect_reply_upcall(ep, status2errno(status));
2282 state_set(&ep->com, DEAD);
2284 if (ep->com.remote_addr.ss_family == AF_INET6) {
2285 struct sockaddr_in6 *sin6 =
2286 (struct sockaddr_in6 *)&ep->com.local_addr;
2287 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2288 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2290 if (status && act_open_has_tid(status))
2291 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2293 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2294 cxgb4_free_atid(t, atid);
2295 dst_release(ep->dst);
2296 cxgb4_l2t_release(ep->l2t);
2297 c4iw_put_ep(&ep->com);
2302 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2304 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2305 unsigned int stid = GET_TID(rpl);
2306 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2309 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2312 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2313 rpl->status, status2errno(rpl->status));
2314 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2315 c4iw_put_ep(&ep->com);
2320 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2322 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2323 unsigned int stid = GET_TID(rpl);
2324 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2326 PDBG("%s ep %p\n", __func__, ep);
2327 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2328 c4iw_put_ep(&ep->com);
2332 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2333 struct cpl_pass_accept_req *req)
2335 struct cpl_pass_accept_rpl *rpl;
2336 unsigned int mtu_idx;
2340 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2342 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2344 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2345 BUG_ON(skb_cloned(skb));
2349 if (!is_t4(adapter_type)) {
2350 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2352 INIT_TP_WR(rpl5, ep->hwtid);
2354 skb_trim(skb, sizeof(*rpl));
2355 INIT_TP_WR(rpl, ep->hwtid);
2357 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2360 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2361 enable_tcp_timestamps && req->tcpopt.tstamp,
2362 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2363 wscale = cxgb_compute_wscale(rcv_win);
2366 * Specify the largest window that will fit in opt0. The
2367 * remainder will be specified in the rx_data_ack.
2369 win = ep->rcv_win >> 10;
2370 if (win > RCV_BUFSIZ_M)
2372 opt0 = (nocong ? NO_CONG_F : 0) |
2375 WND_SCALE_V(wscale) |
2376 MSS_IDX_V(mtu_idx) |
2377 L2T_IDX_V(ep->l2t->idx) |
2378 TX_CHAN_V(ep->tx_chan) |
2379 SMAC_SEL_V(ep->smac_idx) |
2380 DSCP_V(ep->tos >> 2) |
2381 ULP_MODE_V(ULP_MODE_TCPDDP) |
2383 opt2 = RX_CHANNEL_V(0) |
2384 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2386 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2387 opt2 |= TSTAMPS_EN_F;
2388 if (enable_tcp_sack && req->tcpopt.sack)
2390 if (wscale && enable_tcp_window_scaling)
2391 opt2 |= WND_SCALE_EN_F;
2393 const struct tcphdr *tcph;
2394 u32 hlen = ntohl(req->hdr_len);
2396 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2397 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2400 tcph = (const void *)(req + 1) +
2401 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2402 if (tcph->ece && tcph->cwr)
2403 opt2 |= CCTRL_ECN_V(1);
2405 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2406 u32 isn = (prandom_u32() & ~7UL) - 1;
2407 opt2 |= T5_OPT_2_VALID_F;
2408 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2411 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2414 rpl5->iss = cpu_to_be32(isn);
2415 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2418 rpl->opt0 = cpu_to_be64(opt0);
2419 rpl->opt2 = cpu_to_be32(opt2);
2420 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2421 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2423 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2426 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2428 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2429 BUG_ON(skb_cloned(skb));
2430 skb_trim(skb, sizeof(struct cpl_tid_release));
2431 release_tid(&dev->rdev, hwtid, skb);
2435 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2437 struct c4iw_ep *child_ep = NULL, *parent_ep;
2438 struct cpl_pass_accept_req *req = cplhdr(skb);
2439 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2440 struct tid_info *t = dev->rdev.lldi.tids;
2441 unsigned int hwtid = GET_TID(req);
2442 struct dst_entry *dst;
2443 __u8 local_ip[16], peer_ip[16];
2444 __be16 local_port, peer_port;
2445 struct sockaddr_in6 *sin6;
2447 u16 peer_mss = ntohs(req->tcpopt.mss);
2449 unsigned short hdrs;
2450 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2452 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2454 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2458 if (state_read(&parent_ep->com) != LISTEN) {
2459 PDBG("%s - listening ep not in LISTEN\n", __func__);
2463 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2464 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2466 /* Find output route */
2468 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2469 , __func__, parent_ep, hwtid,
2470 local_ip, peer_ip, ntohs(local_port),
2471 ntohs(peer_port), peer_mss);
2472 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2473 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2474 local_port, peer_port, tos);
2476 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2477 , __func__, parent_ep, hwtid,
2478 local_ip, peer_ip, ntohs(local_port),
2479 ntohs(peer_port), peer_mss);
2480 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2481 local_ip, peer_ip, local_port, peer_port,
2482 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2483 ((struct sockaddr_in6 *)
2484 &parent_ep->com.local_addr)->sin6_scope_id);
2487 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2492 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2494 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2500 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2501 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2503 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2510 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2511 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2512 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2513 child_ep->mtu = peer_mss + hdrs;
2515 skb_queue_head_init(&child_ep->com.ep_skb_list);
2516 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2519 state_set(&child_ep->com, CONNECTING);
2520 child_ep->com.dev = dev;
2521 child_ep->com.cm_id = NULL;
2524 struct sockaddr_in *sin = (struct sockaddr_in *)
2525 &child_ep->com.local_addr;
2527 sin->sin_family = PF_INET;
2528 sin->sin_port = local_port;
2529 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2531 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2532 sin->sin_family = PF_INET;
2533 sin->sin_port = ((struct sockaddr_in *)
2534 &parent_ep->com.local_addr)->sin_port;
2535 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2537 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2538 sin->sin_family = PF_INET;
2539 sin->sin_port = peer_port;
2540 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2542 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2543 sin6->sin6_family = PF_INET6;
2544 sin6->sin6_port = local_port;
2545 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2547 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2548 sin6->sin6_family = PF_INET6;
2549 sin6->sin6_port = ((struct sockaddr_in6 *)
2550 &parent_ep->com.local_addr)->sin6_port;
2551 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2553 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2554 sin6->sin6_family = PF_INET6;
2555 sin6->sin6_port = peer_port;
2556 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2559 c4iw_get_ep(&parent_ep->com);
2560 child_ep->parent_ep = parent_ep;
2561 child_ep->tos = tos;
2562 child_ep->dst = dst;
2563 child_ep->hwtid = hwtid;
2565 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2566 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2568 init_timer(&child_ep->timer);
2569 cxgb4_insert_tid(t, child_ep, hwtid);
2570 insert_ep_tid(child_ep);
2571 if (accept_cr(child_ep, skb, req)) {
2572 c4iw_put_ep(&parent_ep->com);
2573 release_ep_resources(child_ep);
2575 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2578 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2579 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2580 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2584 c4iw_put_ep(&child_ep->com);
2586 reject_cr(dev, hwtid, skb);
2589 c4iw_put_ep(&parent_ep->com);
2593 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2596 struct cpl_pass_establish *req = cplhdr(skb);
2597 unsigned int tid = GET_TID(req);
2600 ep = get_ep_from_tid(dev, tid);
2601 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2602 ep->snd_seq = be32_to_cpu(req->snd_isn);
2603 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2605 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2606 ntohs(req->tcp_opt));
2608 set_emss(ep, ntohs(req->tcp_opt));
2610 dst_confirm(ep->dst);
2611 mutex_lock(&ep->com.mutex);
2612 ep->com.state = MPA_REQ_WAIT;
2614 set_bit(PASS_ESTAB, &ep->com.history);
2615 ret = send_flowc(ep);
2616 mutex_unlock(&ep->com.mutex);
2618 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2619 c4iw_put_ep(&ep->com);
2624 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2626 struct cpl_peer_close *hdr = cplhdr(skb);
2628 struct c4iw_qp_attributes attrs;
2631 unsigned int tid = GET_TID(hdr);
2634 ep = get_ep_from_tid(dev, tid);
2638 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2639 dst_confirm(ep->dst);
2641 set_bit(PEER_CLOSE, &ep->com.history);
2642 mutex_lock(&ep->com.mutex);
2643 switch (ep->com.state) {
2645 __state_set(&ep->com, CLOSING);
2648 __state_set(&ep->com, CLOSING);
2649 connect_reply_upcall(ep, -ECONNRESET);
2654 * We're gonna mark this puppy DEAD, but keep
2655 * the reference on it until the ULP accepts or
2656 * rejects the CR. Also wake up anyone waiting
2657 * in rdma connection migration (see c4iw_accept_cr()).
2659 __state_set(&ep->com, CLOSING);
2660 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2661 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2664 __state_set(&ep->com, CLOSING);
2665 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2666 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2670 __state_set(&ep->com, CLOSING);
2671 attrs.next_state = C4IW_QP_STATE_CLOSING;
2672 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2673 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2674 if (ret != -ECONNRESET) {
2675 peer_close_upcall(ep);
2683 __state_set(&ep->com, MORIBUND);
2687 (void)stop_ep_timer(ep);
2688 if (ep->com.cm_id && ep->com.qp) {
2689 attrs.next_state = C4IW_QP_STATE_IDLE;
2690 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2691 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2693 close_complete_upcall(ep, 0);
2694 __state_set(&ep->com, DEAD);
2704 mutex_unlock(&ep->com.mutex);
2706 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2708 release_ep_resources(ep);
2709 c4iw_put_ep(&ep->com);
2713 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2715 struct cpl_abort_req_rss *req = cplhdr(skb);
2717 struct sk_buff *rpl_skb;
2718 struct c4iw_qp_attributes attrs;
2721 unsigned int tid = GET_TID(req);
2722 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2724 ep = get_ep_from_tid(dev, tid);
2728 if (cxgb_is_neg_adv(req->status)) {
2729 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2730 __func__, ep->hwtid, req->status,
2731 neg_adv_str(req->status));
2732 ep->stats.abort_neg_adv++;
2733 mutex_lock(&dev->rdev.stats.lock);
2734 dev->rdev.stats.neg_adv++;
2735 mutex_unlock(&dev->rdev.stats.lock);
2738 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2740 set_bit(PEER_ABORT, &ep->com.history);
2743 * Wake up any threads in rdma_init() or rdma_fini().
2744 * However, this is not needed if com state is just
2747 if (ep->com.state != MPA_REQ_SENT)
2748 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2750 mutex_lock(&ep->com.mutex);
2751 switch (ep->com.state) {
2753 c4iw_put_ep(&ep->parent_ep->com);
2756 (void)stop_ep_timer(ep);
2759 (void)stop_ep_timer(ep);
2760 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2761 connect_reply_upcall(ep, -ECONNRESET);
2764 * we just don't send notification upwards because we
2765 * want to retry with mpa_v1 without upper layers even
2768 * do some housekeeping so as to re-initiate the
2771 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2773 ep->retry_with_mpa_v1 = 1;
2785 if (ep->com.cm_id && ep->com.qp) {
2786 attrs.next_state = C4IW_QP_STATE_ERROR;
2787 ret = c4iw_modify_qp(ep->com.qp->rhp,
2788 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2792 "%s - qp <- error failed!\n",
2795 peer_abort_upcall(ep);
2800 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2801 mutex_unlock(&ep->com.mutex);
2807 dst_confirm(ep->dst);
2808 if (ep->com.state != ABORTING) {
2809 __state_set(&ep->com, DEAD);
2810 /* we don't release if we want to retry with mpa_v1 */
2811 if (!ep->retry_with_mpa_v1)
2814 mutex_unlock(&ep->com.mutex);
2816 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2817 if (WARN_ON(!rpl_skb)) {
2822 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2824 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2827 release_ep_resources(ep);
2828 else if (ep->retry_with_mpa_v1) {
2829 if (ep->com.remote_addr.ss_family == AF_INET6) {
2830 struct sockaddr_in6 *sin6 =
2831 (struct sockaddr_in6 *)
2832 &ep->com.local_addr;
2834 ep->com.dev->rdev.lldi.ports[0],
2835 (const u32 *)&sin6->sin6_addr.s6_addr,
2838 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2839 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2840 dst_release(ep->dst);
2841 cxgb4_l2t_release(ep->l2t);
2846 c4iw_put_ep(&ep->com);
2847 /* Dereferencing ep, referenced in peer_abort_intr() */
2848 c4iw_put_ep(&ep->com);
2852 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2855 struct c4iw_qp_attributes attrs;
2856 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2858 unsigned int tid = GET_TID(rpl);
2860 ep = get_ep_from_tid(dev, tid);
2864 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2867 /* The cm_id may be null if we failed to connect */
2868 mutex_lock(&ep->com.mutex);
2869 set_bit(CLOSE_CON_RPL, &ep->com.history);
2870 switch (ep->com.state) {
2872 __state_set(&ep->com, MORIBUND);
2875 (void)stop_ep_timer(ep);
2876 if ((ep->com.cm_id) && (ep->com.qp)) {
2877 attrs.next_state = C4IW_QP_STATE_IDLE;
2878 c4iw_modify_qp(ep->com.qp->rhp,
2880 C4IW_QP_ATTR_NEXT_STATE,
2883 close_complete_upcall(ep, 0);
2884 __state_set(&ep->com, DEAD);
2894 mutex_unlock(&ep->com.mutex);
2896 release_ep_resources(ep);
2897 c4iw_put_ep(&ep->com);
2901 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2903 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2904 unsigned int tid = GET_TID(rpl);
2906 struct c4iw_qp_attributes attrs;
2908 ep = get_ep_from_tid(dev, tid);
2911 if (ep && ep->com.qp) {
2912 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2913 ep->com.qp->wq.sq.qid);
2914 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2915 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2916 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2918 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2919 c4iw_put_ep(&ep->com);
2925 * Upcall from the adapter indicating data has been transmitted.
2926 * For us its just the single MPA request or reply. We can now free
2927 * the skb holding the mpa message.
2929 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2932 struct cpl_fw4_ack *hdr = cplhdr(skb);
2933 u8 credits = hdr->credits;
2934 unsigned int tid = GET_TID(hdr);
2937 ep = get_ep_from_tid(dev, tid);
2940 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2942 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2943 __func__, ep, ep->hwtid, state_read(&ep->com));
2947 dst_confirm(ep->dst);
2949 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2950 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2951 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2952 mutex_lock(&ep->com.mutex);
2953 kfree_skb(ep->mpa_skb);
2955 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2957 mutex_unlock(&ep->com.mutex);
2960 c4iw_put_ep(&ep->com);
2964 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2967 struct c4iw_ep *ep = to_ep(cm_id);
2969 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2971 mutex_lock(&ep->com.mutex);
2972 if (ep->com.state != MPA_REQ_RCVD) {
2973 mutex_unlock(&ep->com.mutex);
2974 c4iw_put_ep(&ep->com);
2977 set_bit(ULP_REJECT, &ep->com.history);
2981 abort = send_mpa_reject(ep, pdata, pdata_len);
2982 mutex_unlock(&ep->com.mutex);
2985 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
2986 c4iw_put_ep(&ep->com);
2990 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2993 struct c4iw_qp_attributes attrs;
2994 enum c4iw_qp_attr_mask mask;
2995 struct c4iw_ep *ep = to_ep(cm_id);
2996 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2997 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3000 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3002 mutex_lock(&ep->com.mutex);
3003 if (ep->com.state != MPA_REQ_RCVD) {
3010 set_bit(ULP_ACCEPT, &ep->com.history);
3011 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3012 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3017 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3018 if (conn_param->ord > ep->ird) {
3019 if (RELAXED_IRD_NEGOTIATION) {
3020 conn_param->ord = ep->ird;
3022 ep->ird = conn_param->ird;
3023 ep->ord = conn_param->ord;
3024 send_mpa_reject(ep, conn_param->private_data,
3025 conn_param->private_data_len);
3030 if (conn_param->ird < ep->ord) {
3031 if (RELAXED_IRD_NEGOTIATION &&
3032 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3033 conn_param->ird = ep->ord;
3040 ep->ird = conn_param->ird;
3041 ep->ord = conn_param->ord;
3043 if (ep->mpa_attr.version == 1) {
3044 if (peer2peer && ep->ird == 0)
3048 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3049 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3053 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3055 ep->com.cm_id = cm_id;
3056 ref_cm_id(&ep->com);
3060 /* bind QP to EP and move to RTS */
3061 attrs.mpa_attr = ep->mpa_attr;
3062 attrs.max_ird = ep->ird;
3063 attrs.max_ord = ep->ord;
3064 attrs.llp_stream_handle = ep;
3065 attrs.next_state = C4IW_QP_STATE_RTS;
3067 /* bind QP and TID with INIT_WR */
3068 mask = C4IW_QP_ATTR_NEXT_STATE |
3069 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3070 C4IW_QP_ATTR_MPA_ATTR |
3071 C4IW_QP_ATTR_MAX_IRD |
3072 C4IW_QP_ATTR_MAX_ORD;
3074 err = c4iw_modify_qp(ep->com.qp->rhp,
3075 ep->com.qp, mask, &attrs, 1);
3077 goto err_deref_cm_id;
3079 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3080 err = send_mpa_reply(ep, conn_param->private_data,
3081 conn_param->private_data_len);
3083 goto err_deref_cm_id;
3085 __state_set(&ep->com, FPDU_MODE);
3086 established_upcall(ep);
3087 mutex_unlock(&ep->com.mutex);
3088 c4iw_put_ep(&ep->com);
3091 deref_cm_id(&ep->com);
3095 mutex_unlock(&ep->com.mutex);
3097 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3098 c4iw_put_ep(&ep->com);
3102 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3104 struct in_device *ind;
3106 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3107 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3109 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3111 return -EADDRNOTAVAIL;
3112 for_primary_ifa(ind) {
3113 laddr->sin_addr.s_addr = ifa->ifa_address;
3114 raddr->sin_addr.s_addr = ifa->ifa_address;
3120 return found ? 0 : -EADDRNOTAVAIL;
3123 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3124 unsigned char banned_flags)
3126 struct inet6_dev *idev;
3127 int err = -EADDRNOTAVAIL;
3130 idev = __in6_dev_get(dev);
3132 struct inet6_ifaddr *ifp;
3134 read_lock_bh(&idev->lock);
3135 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3136 if (ifp->scope == IFA_LINK &&
3137 !(ifp->flags & banned_flags)) {
3138 memcpy(addr, &ifp->addr, 16);
3143 read_unlock_bh(&idev->lock);
3149 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3151 struct in6_addr uninitialized_var(addr);
3152 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3153 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3155 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3156 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3157 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3160 return -EADDRNOTAVAIL;
3163 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3165 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3168 struct sockaddr_in *laddr;
3169 struct sockaddr_in *raddr;
3170 struct sockaddr_in6 *laddr6;
3171 struct sockaddr_in6 *raddr6;
3175 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3176 (conn_param->ird > cur_max_read_depth(dev))) {
3180 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3182 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3187 skb_queue_head_init(&ep->com.ep_skb_list);
3188 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3193 init_timer(&ep->timer);
3194 ep->plen = conn_param->private_data_len;
3196 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3197 conn_param->private_data, ep->plen);
3198 ep->ird = conn_param->ird;
3199 ep->ord = conn_param->ord;
3201 if (peer2peer && ep->ord == 0)
3204 ep->com.cm_id = cm_id;
3205 ref_cm_id(&ep->com);
3207 ep->com.qp = get_qhp(dev, conn_param->qpn);
3209 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3214 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3218 * Allocate an active TID to initiate a TCP connection.
3220 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3221 if (ep->atid == -1) {
3222 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3226 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3228 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3229 sizeof(ep->com.local_addr));
3230 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3231 sizeof(ep->com.remote_addr));
3233 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3234 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3235 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3236 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3238 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3240 ra = (__u8 *)&raddr->sin_addr;
3243 * Handle loopback requests to INADDR_ANY.
3245 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3246 err = pick_local_ipaddrs(dev, cm_id);
3252 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3253 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3254 ra, ntohs(raddr->sin_port));
3255 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3256 laddr->sin_addr.s_addr,
3257 raddr->sin_addr.s_addr,
3259 raddr->sin_port, cm_id->tos);
3262 ra = (__u8 *)&raddr6->sin6_addr;
3265 * Handle loopback requests to INADDR_ANY.
3267 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3268 err = pick_local_ip6addrs(dev, cm_id);
3274 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3275 __func__, laddr6->sin6_addr.s6_addr,
3276 ntohs(laddr6->sin6_port),
3277 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3278 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3279 laddr6->sin6_addr.s6_addr,
3280 raddr6->sin6_addr.s6_addr,
3282 raddr6->sin6_port, cm_id->tos,
3283 raddr6->sin6_scope_id);
3286 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3287 err = -EHOSTUNREACH;
3291 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3292 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3294 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3298 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3299 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3302 state_set(&ep->com, CONNECTING);
3303 ep->tos = cm_id->tos;
3305 /* send connect request to rnic */
3306 err = send_connect(ep);
3310 cxgb4_l2t_release(ep->l2t);
3312 dst_release(ep->dst);
3314 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3315 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3317 skb_queue_purge(&ep->com.ep_skb_list);
3318 deref_cm_id(&ep->com);
3320 c4iw_put_ep(&ep->com);
3325 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3328 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3329 &ep->com.local_addr;
3331 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3332 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3333 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3337 c4iw_init_wr_wait(&ep->com.wr_wait);
3338 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3339 ep->stid, &sin6->sin6_addr,
3341 ep->com.dev->rdev.lldi.rxq_ids[0]);
3343 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3347 err = net_xmit_errno(err);
3349 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3350 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3351 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3353 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3358 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3361 struct sockaddr_in *sin = (struct sockaddr_in *)
3362 &ep->com.local_addr;
3364 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3366 err = cxgb4_create_server_filter(
3367 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3368 sin->sin_addr.s_addr, sin->sin_port, 0,
3369 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3370 if (err == -EBUSY) {
3371 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3375 set_current_state(TASK_UNINTERRUPTIBLE);
3376 schedule_timeout(usecs_to_jiffies(100));
3378 } while (err == -EBUSY);
3380 c4iw_init_wr_wait(&ep->com.wr_wait);
3381 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3382 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3383 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3385 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3389 err = net_xmit_errno(err);
3392 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3394 &sin->sin_addr, ntohs(sin->sin_port));
3398 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3401 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3402 struct c4iw_listen_ep *ep;
3406 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3408 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3412 skb_queue_head_init(&ep->com.ep_skb_list);
3413 PDBG("%s ep %p\n", __func__, ep);
3414 ep->com.cm_id = cm_id;
3415 ref_cm_id(&ep->com);
3417 ep->backlog = backlog;
3418 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3419 sizeof(ep->com.local_addr));
3422 * Allocate a server TID.
3424 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3425 ep->com.local_addr.ss_family == AF_INET)
3426 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3427 cm_id->m_local_addr.ss_family, ep);
3429 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3430 cm_id->m_local_addr.ss_family, ep);
3432 if (ep->stid == -1) {
3433 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3437 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3439 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3440 sizeof(ep->com.local_addr));
3442 state_set(&ep->com, LISTEN);
3443 if (ep->com.local_addr.ss_family == AF_INET)
3444 err = create_server4(dev, ep);
3446 err = create_server6(dev, ep);
3448 cm_id->provider_data = ep;
3451 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3452 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3453 ep->com.local_addr.ss_family);
3455 deref_cm_id(&ep->com);
3456 c4iw_put_ep(&ep->com);
3462 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3465 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3467 PDBG("%s ep %p\n", __func__, ep);
3470 state_set(&ep->com, DEAD);
3471 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3472 ep->com.local_addr.ss_family == AF_INET) {
3473 err = cxgb4_remove_server_filter(
3474 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3475 ep->com.dev->rdev.lldi.rxq_ids[0], false);
3477 struct sockaddr_in6 *sin6;
3478 c4iw_init_wr_wait(&ep->com.wr_wait);
3479 err = cxgb4_remove_server(
3480 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3481 ep->com.dev->rdev.lldi.rxq_ids[0],
3482 ep->com.local_addr.ss_family == AF_INET6);
3485 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3487 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3488 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3489 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3491 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3492 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3493 ep->com.local_addr.ss_family);
3495 deref_cm_id(&ep->com);
3496 c4iw_put_ep(&ep->com);
3500 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3505 struct c4iw_rdev *rdev;
3507 mutex_lock(&ep->com.mutex);
3509 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3510 states[ep->com.state], abrupt);
3513 * Ref the ep here in case we have fatal errors causing the
3514 * ep to be released and freed.
3516 c4iw_get_ep(&ep->com);
3518 rdev = &ep->com.dev->rdev;
3519 if (c4iw_fatal_error(rdev)) {
3521 close_complete_upcall(ep, -EIO);
3522 ep->com.state = DEAD;
3524 switch (ep->com.state) {
3533 ep->com.state = ABORTING;
3535 ep->com.state = CLOSING;
3538 * if we close before we see the fw4_ack() then we fix
3539 * up the timer state since we're reusing it.
3542 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3543 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3548 set_bit(CLOSE_SENT, &ep->com.flags);
3551 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3554 (void)stop_ep_timer(ep);
3555 ep->com.state = ABORTING;
3557 ep->com.state = MORIBUND;
3563 PDBG("%s ignoring disconnect ep %p state %u\n",
3564 __func__, ep, ep->com.state);
3573 set_bit(EP_DISC_ABORT, &ep->com.history);
3574 ret = send_abort(ep);
3576 set_bit(EP_DISC_CLOSE, &ep->com.history);
3577 ret = send_halfclose(ep);
3580 set_bit(EP_DISC_FAIL, &ep->com.history);
3583 close_complete_upcall(ep, -EIO);
3586 struct c4iw_qp_attributes attrs;
3588 attrs.next_state = C4IW_QP_STATE_ERROR;
3589 ret = c4iw_modify_qp(ep->com.qp->rhp,
3591 C4IW_QP_ATTR_NEXT_STATE,
3595 "%s - qp <- error failed!\n",
3601 mutex_unlock(&ep->com.mutex);
3602 c4iw_put_ep(&ep->com);
3604 release_ep_resources(ep);
3608 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3609 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3612 int atid = be32_to_cpu(req->tid);
3614 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3615 (__force u32) req->tid);
3619 switch (req->retval) {
3621 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3622 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3623 send_fw_act_open_req(ep, atid);
3627 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3628 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3629 send_fw_act_open_req(ep, atid);
3634 pr_info("%s unexpected ofld conn wr retval %d\n",
3635 __func__, req->retval);
3638 pr_err("active ofld_connect_wr failure %d atid %d\n",
3640 mutex_lock(&dev->rdev.stats.lock);
3641 dev->rdev.stats.act_ofld_conn_fails++;
3642 mutex_unlock(&dev->rdev.stats.lock);
3643 connect_reply_upcall(ep, status2errno(req->retval));
3644 state_set(&ep->com, DEAD);
3645 if (ep->com.remote_addr.ss_family == AF_INET6) {
3646 struct sockaddr_in6 *sin6 =
3647 (struct sockaddr_in6 *)&ep->com.local_addr;
3648 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3649 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3651 remove_handle(dev, &dev->atid_idr, atid);
3652 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3653 dst_release(ep->dst);
3654 cxgb4_l2t_release(ep->l2t);
3655 c4iw_put_ep(&ep->com);
3658 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3659 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3661 struct sk_buff *rpl_skb;
3662 struct cpl_pass_accept_req *cpl;
3665 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3668 PDBG("%s passive open failure %d\n", __func__, req->retval);
3669 mutex_lock(&dev->rdev.stats.lock);
3670 dev->rdev.stats.pas_ofld_conn_fails++;
3671 mutex_unlock(&dev->rdev.stats.lock);
3674 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3675 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3676 (__force u32) htonl(
3677 (__force u32) req->tid)));
3678 ret = pass_accept_req(dev, rpl_skb);
3685 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3687 struct cpl_fw6_msg *rpl = cplhdr(skb);
3688 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3690 switch (rpl->type) {
3692 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3694 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3695 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3696 switch (req->t_state) {
3698 active_ofld_conn_reply(dev, skb, req);
3701 passive_ofld_conn_reply(dev, skb, req);
3704 pr_err("%s unexpected ofld conn wr state %d\n",
3705 __func__, req->t_state);
3713 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3716 __be16 hdr_len, vlantag, len;
3718 int tcp_hdr_len, ip_hdr_len;
3720 struct cpl_rx_pkt *cpl = cplhdr(skb);
3721 struct cpl_pass_accept_req *req;
3722 struct tcp_options_received tmp_opt;
3723 struct c4iw_dev *dev;
3724 enum chip_type type;
3726 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3727 /* Store values from cpl_rx_pkt in temporary location. */
3728 vlantag = cpl->vlan;
3730 l2info = cpl->l2info;
3731 hdr_len = cpl->hdr_len;
3734 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3737 * We need to parse the TCP options from SYN packet.
3738 * to generate cpl_pass_accept_req.
3740 memset(&tmp_opt, 0, sizeof(tmp_opt));
3741 tcp_clear_options(&tmp_opt);
3742 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3744 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3745 memset(req, 0, sizeof(*req));
3746 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3747 SYN_MAC_IDX_V(RX_MACIDX_G(
3748 be32_to_cpu(l2info))) |
3750 type = dev->rdev.lldi.adapter_type;
3751 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3752 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3754 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3755 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3756 eth_hdr_len = is_t4(type) ?
3757 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3758 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3759 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3760 IP_HDR_LEN_V(ip_hdr_len) |
3761 ETH_HDR_LEN_V(eth_hdr_len));
3762 } else { /* T6 and later */
3763 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3764 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3765 T6_IP_HDR_LEN_V(ip_hdr_len) |
3766 T6_ETH_HDR_LEN_V(eth_hdr_len));
3768 req->vlan = vlantag;
3770 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3771 PASS_OPEN_TOS_V(tos));
3772 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3773 if (tmp_opt.wscale_ok)
3774 req->tcpopt.wsf = tmp_opt.snd_wscale;
3775 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3776 if (tmp_opt.sack_ok)
3777 req->tcpopt.sack = 1;
3778 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3782 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3783 __be32 laddr, __be16 lport,
3784 __be32 raddr, __be16 rport,
3785 u32 rcv_isn, u32 filter, u16 window,
3786 u32 rss_qid, u8 port_id)
3788 struct sk_buff *req_skb;
3789 struct fw_ofld_connection_wr *req;
3790 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3793 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3794 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3795 memset(req, 0, sizeof(*req));
3796 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3797 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3798 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3799 req->le.filter = (__force __be32) filter;
3800 req->le.lport = lport;
3801 req->le.pport = rport;
3802 req->le.u.ipv4.lip = laddr;
3803 req->le.u.ipv4.pip = raddr;
3804 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3805 req->tcb.rcv_adv = htons(window);
3806 req->tcb.t_state_to_astid =
3807 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3808 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3809 FW_OFLD_CONNECTION_WR_ASTID_V(
3810 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3813 * We store the qid in opt2 which will be used by the firmware
3814 * to send us the wr response.
3816 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3819 * We initialize the MSS index in TCB to 0xF.
3820 * So that when driver sends cpl_pass_accept_rpl
3821 * TCB picks up the correct value. If this was 0
3822 * TP will ignore any value > 0 for MSS index.
3824 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3825 req->cookie = (uintptr_t)skb;
3827 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3828 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3830 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3838 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3839 * messages when a filter is being used instead of server to
3840 * redirect a syn packet. When packets hit filter they are redirected
3841 * to the offload queue and driver tries to establish the connection
3842 * using firmware work request.
3844 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3847 unsigned int filter;
3848 struct ethhdr *eh = NULL;
3849 struct vlan_ethhdr *vlan_eh = NULL;
3851 struct tcphdr *tcph;
3852 struct rss_header *rss = (void *)skb->data;
3853 struct cpl_rx_pkt *cpl = (void *)skb->data;
3854 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3855 struct l2t_entry *e;
3856 struct dst_entry *dst;
3857 struct c4iw_ep *lep = NULL;
3859 struct port_info *pi;
3860 struct net_device *pdev;
3861 u16 rss_qid, eth_hdr_len;
3864 struct neighbour *neigh;
3866 /* Drop all non-SYN packets */
3867 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3871 * Drop all packets which did not hit the filter.
3872 * Unlikely to happen.
3874 if (!(rss->filter_hit && rss->filter_tid))
3878 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3880 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3882 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3884 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3888 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3890 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3893 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3896 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3899 pr_err("T%d Chip is not supported\n",
3900 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3904 if (eth_hdr_len == ETH_HLEN) {
3905 eh = (struct ethhdr *)(req + 1);
3906 iph = (struct iphdr *)(eh + 1);
3908 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3909 iph = (struct iphdr *)(vlan_eh + 1);
3910 skb->vlan_tci = ntohs(cpl->vlan);
3913 if (iph->version != 0x4)
3916 tcph = (struct tcphdr *)(iph + 1);
3917 skb_set_network_header(skb, (void *)iph - (void *)rss);
3918 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3921 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3922 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3923 ntohs(tcph->source), iph->tos);
3925 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3926 iph->daddr, iph->saddr, tcph->dest,
3927 tcph->source, iph->tos);
3929 pr_err("%s - failed to find dst entry!\n",
3933 neigh = dst_neigh_lookup_skb(dst, skb);
3936 pr_err("%s - failed to allocate neigh!\n",
3941 if (neigh->dev->flags & IFF_LOOPBACK) {
3942 pdev = ip_dev_find(&init_net, iph->daddr);
3943 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3945 pi = (struct port_info *)netdev_priv(pdev);
3946 tx_chan = cxgb4_port_chan(pdev);
3949 pdev = get_real_dev(neigh->dev);
3950 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3952 pi = (struct port_info *)netdev_priv(pdev);
3953 tx_chan = cxgb4_port_chan(pdev);
3955 neigh_release(neigh);
3957 pr_err("%s - failed to allocate l2t entry!\n",
3962 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3963 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3964 window = (__force u16) htons((__force u16)tcph->window);
3966 /* Calcuate filter portion for LE region. */
3967 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3968 dev->rdev.lldi.ports[0],
3972 * Synthesize the cpl_pass_accept_req. We have everything except the
3973 * TID. Once firmware sends a reply with TID we update the TID field
3974 * in cpl and pass it through the regular cpl_pass_accept_req path.
3976 build_cpl_pass_accept_req(skb, stid, iph->tos);
3977 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3978 tcph->source, ntohl(tcph->seq), filter, window,
3979 rss_qid, pi->port_id);
3980 cxgb4_l2t_release(e);
3985 c4iw_put_ep(&lep->com);
3990 * These are the real handlers that are called from a
3993 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
3994 [CPL_ACT_ESTABLISH] = act_establish,
3995 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3996 [CPL_RX_DATA] = rx_data,
3997 [CPL_ABORT_RPL_RSS] = abort_rpl,
3998 [CPL_ABORT_RPL] = abort_rpl,
3999 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4000 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4001 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4002 [CPL_PASS_ESTABLISH] = pass_establish,
4003 [CPL_PEER_CLOSE] = peer_close,
4004 [CPL_ABORT_REQ_RSS] = peer_abort,
4005 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4006 [CPL_RDMA_TERMINATE] = terminate,
4007 [CPL_FW4_ACK] = fw4_ack,
4008 [CPL_FW6_MSG] = deferred_fw6_msg,
4009 [CPL_RX_PKT] = rx_pkt,
4010 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4011 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4014 static void process_timeout(struct c4iw_ep *ep)
4016 struct c4iw_qp_attributes attrs;
4019 mutex_lock(&ep->com.mutex);
4020 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4022 set_bit(TIMEDOUT, &ep->com.history);
4023 switch (ep->com.state) {
4025 connect_reply_upcall(ep, -ETIMEDOUT);
4034 if (ep->com.cm_id && ep->com.qp) {
4035 attrs.next_state = C4IW_QP_STATE_ERROR;
4036 c4iw_modify_qp(ep->com.qp->rhp,
4037 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4040 close_complete_upcall(ep, -ETIMEDOUT);
4046 * These states are expected if the ep timed out at the same
4047 * time as another thread was calling stop_ep_timer().
4048 * So we silently do nothing for these states.
4053 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4054 __func__, ep, ep->hwtid, ep->com.state);
4057 mutex_unlock(&ep->com.mutex);
4059 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4060 c4iw_put_ep(&ep->com);
4063 static void process_timedout_eps(void)
4067 spin_lock_irq(&timeout_lock);
4068 while (!list_empty(&timeout_list)) {
4069 struct list_head *tmp;
4071 tmp = timeout_list.next;
4075 spin_unlock_irq(&timeout_lock);
4076 ep = list_entry(tmp, struct c4iw_ep, entry);
4077 process_timeout(ep);
4078 spin_lock_irq(&timeout_lock);
4080 spin_unlock_irq(&timeout_lock);
4083 static void process_work(struct work_struct *work)
4085 struct sk_buff *skb = NULL;
4086 struct c4iw_dev *dev;
4087 struct cpl_act_establish *rpl;
4088 unsigned int opcode;
4091 process_timedout_eps();
4092 while ((skb = skb_dequeue(&rxq))) {
4094 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4095 opcode = rpl->ot.opcode;
4097 BUG_ON(!work_handlers[opcode]);
4098 ret = work_handlers[opcode](dev, skb);
4101 process_timedout_eps();
4105 static DECLARE_WORK(skb_work, process_work);
4107 static void ep_timeout(unsigned long arg)
4109 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4112 spin_lock(&timeout_lock);
4113 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4115 * Only insert if it is not already on the list.
4117 if (!ep->entry.next) {
4118 list_add_tail(&ep->entry, &timeout_list);
4122 spin_unlock(&timeout_lock);
4124 queue_work(workq, &skb_work);
4128 * All the CM events are handled on a work queue to have a safe context.
4130 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4134 * Save dev in the skb->cb area.
4136 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4139 * Queue the skb and schedule the worker thread.
4141 skb_queue_tail(&rxq, skb);
4142 queue_work(workq, &skb_work);
4146 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4148 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4150 if (rpl->status != CPL_ERR_NONE) {
4151 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4152 "for tid %u\n", rpl->status, GET_TID(rpl));
4158 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4160 struct cpl_fw6_msg *rpl = cplhdr(skb);
4161 struct c4iw_wr_wait *wr_waitp;
4164 PDBG("%s type %u\n", __func__, rpl->type);
4166 switch (rpl->type) {
4167 case FW6_TYPE_WR_RPL:
4168 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4169 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4170 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4172 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4176 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4180 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4188 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4190 struct cpl_abort_req_rss *req = cplhdr(skb);
4192 unsigned int tid = GET_TID(req);
4194 ep = get_ep_from_tid(dev, tid);
4195 /* This EP will be dereferenced in peer_abort() */
4197 printk(KERN_WARNING MOD
4198 "Abort on non-existent endpoint, tid %d\n", tid);
4202 if (cxgb_is_neg_adv(req->status)) {
4203 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4204 __func__, ep->hwtid, req->status,
4205 neg_adv_str(req->status));
4208 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4211 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4218 * Most upcalls from the T4 Core go to sched() to
4219 * schedule the processing on a work queue.
4221 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4222 [CPL_ACT_ESTABLISH] = sched,
4223 [CPL_ACT_OPEN_RPL] = sched,
4224 [CPL_RX_DATA] = sched,
4225 [CPL_ABORT_RPL_RSS] = sched,
4226 [CPL_ABORT_RPL] = sched,
4227 [CPL_PASS_OPEN_RPL] = sched,
4228 [CPL_CLOSE_LISTSRV_RPL] = sched,
4229 [CPL_PASS_ACCEPT_REQ] = sched,
4230 [CPL_PASS_ESTABLISH] = sched,
4231 [CPL_PEER_CLOSE] = sched,
4232 [CPL_CLOSE_CON_RPL] = sched,
4233 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4234 [CPL_RDMA_TERMINATE] = sched,
4235 [CPL_FW4_ACK] = sched,
4236 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4237 [CPL_FW6_MSG] = fw6_msg,
4238 [CPL_RX_PKT] = sched
4241 int __init c4iw_cm_init(void)
4243 spin_lock_init(&timeout_lock);
4244 skb_queue_head_init(&rxq);
4246 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4253 void c4iw_cm_term(void)
4255 WARN_ON(!list_empty(&timeout_list));
4256 flush_workqueue(workq);
4257 destroy_workqueue(workq);