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 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
338 static void insert_ep_tid(struct c4iw_ep *ep)
342 spin_lock_irqsave(&ep->com.dev->lock, flags);
343 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
344 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
348 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
350 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
355 spin_lock_irqsave(&dev->lock, flags);
356 ep = idr_find(&dev->hwtid_idr, tid);
358 c4iw_get_ep(&ep->com);
359 spin_unlock_irqrestore(&dev->lock, flags);
364 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
366 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
369 struct c4iw_listen_ep *ep;
372 spin_lock_irqsave(&dev->lock, flags);
373 ep = idr_find(&dev->stid_idr, stid);
375 c4iw_get_ep(&ep->com);
376 spin_unlock_irqrestore(&dev->lock, flags);
380 void _c4iw_free_ep(struct kref *kref)
384 ep = container_of(kref, struct c4iw_ep, com.kref);
385 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
386 if (test_bit(QP_REFERENCED, &ep->com.flags))
388 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
389 if (ep->com.remote_addr.ss_family == AF_INET6) {
390 struct sockaddr_in6 *sin6 =
391 (struct sockaddr_in6 *)
395 ep->com.dev->rdev.lldi.ports[0],
396 (const u32 *)&sin6->sin6_addr.s6_addr,
399 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
400 dst_release(ep->dst);
401 cxgb4_l2t_release(ep->l2t);
403 kfree_skb(ep->mpa_skb);
405 if (!skb_queue_empty(&ep->com.ep_skb_list))
406 skb_queue_purge(&ep->com.ep_skb_list);
410 static void release_ep_resources(struct c4iw_ep *ep)
412 set_bit(RELEASE_RESOURCES, &ep->com.flags);
415 * If we have a hwtid, then remove it from the idr table
416 * so lookups will no longer find this endpoint. Otherwise
417 * we have a race where one thread finds the ep ptr just
418 * before the other thread is freeing the ep memory.
422 c4iw_put_ep(&ep->com);
425 static int status2errno(int status)
430 case CPL_ERR_CONN_RESET:
432 case CPL_ERR_ARP_MISS:
433 return -EHOSTUNREACH;
434 case CPL_ERR_CONN_TIMEDOUT:
436 case CPL_ERR_TCAM_FULL:
438 case CPL_ERR_CONN_EXIST:
446 * Try and reuse skbs already allocated...
448 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
450 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
453 skb_reset_transport_header(skb);
455 skb = alloc_skb(len, gfp);
457 t4_set_arp_err_handler(skb, NULL, NULL);
461 static struct net_device *get_real_dev(struct net_device *egress_dev)
463 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
466 static void arp_failure_discard(void *handle, struct sk_buff *skb)
468 pr_err(MOD "ARP failure\n");
472 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
474 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
479 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
480 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
483 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
487 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
488 release_ep_resources(ep);
492 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
496 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
497 c4iw_put_ep(&ep->parent_ep->com);
498 release_ep_resources(ep);
503 * Fake up a special CPL opcode and call sched() so process_work() will call
504 * _put_ep_safe() in a safe context to free the ep resources. This is needed
505 * because ARP error handlers are called in an ATOMIC context, and
506 * _c4iw_free_ep() needs to block.
508 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
511 struct cpl_act_establish *rpl = cplhdr(skb);
513 /* Set our special ARP_FAILURE opcode */
514 rpl->ot.opcode = cpl;
517 * Save ep in the skb->cb area, after where sched() will save the dev
520 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
521 sched(ep->com.dev, skb);
524 /* Handle an ARP failure for an accept */
525 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
527 struct c4iw_ep *ep = handle;
529 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
532 __state_set(&ep->com, DEAD);
533 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
537 * Handle an ARP failure for an active open.
539 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
541 struct c4iw_ep *ep = handle;
543 printk(KERN_ERR MOD "ARP failure during connect\n");
544 connect_reply_upcall(ep, -EHOSTUNREACH);
545 __state_set(&ep->com, DEAD);
546 if (ep->com.remote_addr.ss_family == AF_INET6) {
547 struct sockaddr_in6 *sin6 =
548 (struct sockaddr_in6 *)&ep->com.local_addr;
549 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
550 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
552 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
553 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
554 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
558 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
561 static void abort_arp_failure(void *handle, struct sk_buff *skb)
564 struct c4iw_ep *ep = handle;
565 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
566 struct cpl_abort_req *req = cplhdr(skb);
568 PDBG("%s rdev %p\n", __func__, rdev);
569 req->cmd = CPL_ABORT_NO_RST;
570 ret = c4iw_ofld_send(rdev, skb);
572 __state_set(&ep->com, DEAD);
573 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
577 static int send_flowc(struct c4iw_ep *ep)
579 struct fw_flowc_wr *flowc;
580 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
582 u16 vlan = ep->l2t->vlan;
588 if (vlan == CPL_L2T_VLAN_NONE)
593 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
595 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
596 FW_FLOWC_WR_NPARAMS_V(nparams));
597 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
598 16)) | FW_WR_FLOWID_V(ep->hwtid));
600 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
601 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
602 (ep->com.dev->rdev.lldi.pf));
603 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
604 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
605 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
606 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
607 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
608 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
609 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
610 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
611 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
612 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
613 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
614 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
615 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
616 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
620 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
621 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
622 flowc->mnemval[8].val = cpu_to_be32(pri);
624 /* Pad WR to 16 byte boundary */
625 flowc->mnemval[8].mnemonic = 0;
626 flowc->mnemval[8].val = 0;
628 for (i = 0; i < 9; i++) {
629 flowc->mnemval[i].r4[0] = 0;
630 flowc->mnemval[i].r4[1] = 0;
631 flowc->mnemval[i].r4[2] = 0;
634 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
635 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
638 static int send_halfclose(struct c4iw_ep *ep)
640 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
641 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
643 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
647 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
648 NULL, arp_failure_discard);
650 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
653 static int send_abort(struct c4iw_ep *ep)
655 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
656 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
658 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
659 if (WARN_ON(!req_skb))
662 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
663 ep, abort_arp_failure);
665 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
668 static int send_connect(struct c4iw_ep *ep)
670 struct cpl_act_open_req *req = NULL;
671 struct cpl_t5_act_open_req *t5req = NULL;
672 struct cpl_t6_act_open_req *t6req = NULL;
673 struct cpl_act_open_req6 *req6 = NULL;
674 struct cpl_t5_act_open_req6 *t5req6 = NULL;
675 struct cpl_t6_act_open_req6 *t6req6 = NULL;
679 unsigned int mtu_idx;
681 int win, sizev4, sizev6, wrlen;
682 struct sockaddr_in *la = (struct sockaddr_in *)
684 struct sockaddr_in *ra = (struct sockaddr_in *)
685 &ep->com.remote_addr;
686 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
688 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
689 &ep->com.remote_addr;
691 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
692 u32 isn = (prandom_u32() & ~7UL) - 1;
694 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
696 sizev4 = sizeof(struct cpl_act_open_req);
697 sizev6 = sizeof(struct cpl_act_open_req6);
700 sizev4 = sizeof(struct cpl_t5_act_open_req);
701 sizev6 = sizeof(struct cpl_t5_act_open_req6);
704 sizev4 = sizeof(struct cpl_t6_act_open_req);
705 sizev6 = sizeof(struct cpl_t6_act_open_req6);
708 pr_err("T%d Chip is not supported\n",
709 CHELSIO_CHIP_VERSION(adapter_type));
713 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
714 roundup(sizev4, 16) :
717 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
719 skb = get_skb(NULL, wrlen, GFP_KERNEL);
721 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
725 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
727 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
728 enable_tcp_timestamps,
729 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
730 wscale = cxgb_compute_wscale(rcv_win);
733 * Specify the largest window that will fit in opt0. The
734 * remainder will be specified in the rx_data_ack.
736 win = ep->rcv_win >> 10;
737 if (win > RCV_BUFSIZ_M)
740 opt0 = (nocong ? NO_CONG_F : 0) |
743 WND_SCALE_V(wscale) |
745 L2T_IDX_V(ep->l2t->idx) |
746 TX_CHAN_V(ep->tx_chan) |
747 SMAC_SEL_V(ep->smac_idx) |
748 DSCP_V(ep->tos >> 2) |
749 ULP_MODE_V(ULP_MODE_TCPDDP) |
751 opt2 = RX_CHANNEL_V(0) |
752 CCTRL_ECN_V(enable_ecn) |
753 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
754 if (enable_tcp_timestamps)
755 opt2 |= TSTAMPS_EN_F;
758 if (wscale && enable_tcp_window_scaling)
759 opt2 |= WND_SCALE_EN_F;
760 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
764 opt2 |= T5_OPT_2_VALID_F;
765 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
769 if (ep->com.remote_addr.ss_family == AF_INET6)
770 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
771 (const u32 *)&la6->sin6_addr.s6_addr, 1);
773 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
775 if (ep->com.remote_addr.ss_family == AF_INET) {
776 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
778 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
782 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
784 INIT_TP_WR(t5req, 0);
785 req = (struct cpl_act_open_req *)t5req;
788 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
790 INIT_TP_WR(t6req, 0);
791 req = (struct cpl_act_open_req *)t6req;
792 t5req = (struct cpl_t5_act_open_req *)t6req;
795 pr_err("T%d Chip is not supported\n",
796 CHELSIO_CHIP_VERSION(adapter_type));
801 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
802 ((ep->rss_qid<<14) | ep->atid)));
803 req->local_port = la->sin_port;
804 req->peer_port = ra->sin_port;
805 req->local_ip = la->sin_addr.s_addr;
806 req->peer_ip = ra->sin_addr.s_addr;
807 req->opt0 = cpu_to_be64(opt0);
809 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
810 req->params = cpu_to_be32(cxgb4_select_ntuple(
811 ep->com.dev->rdev.lldi.ports[0],
813 req->opt2 = cpu_to_be32(opt2);
815 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
817 ep->com.dev->rdev.lldi.ports[0],
819 t5req->rsvd = cpu_to_be32(isn);
820 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
821 t5req->opt2 = cpu_to_be32(opt2);
824 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
826 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
830 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
832 INIT_TP_WR(t5req6, 0);
833 req6 = (struct cpl_act_open_req6 *)t5req6;
836 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
838 INIT_TP_WR(t6req6, 0);
839 req6 = (struct cpl_act_open_req6 *)t6req6;
840 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
843 pr_err("T%d Chip is not supported\n",
844 CHELSIO_CHIP_VERSION(adapter_type));
849 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
850 ((ep->rss_qid<<14)|ep->atid)));
851 req6->local_port = la6->sin6_port;
852 req6->peer_port = ra6->sin6_port;
853 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
854 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
855 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
856 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
857 req6->opt0 = cpu_to_be64(opt0);
859 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
860 req6->params = cpu_to_be32(cxgb4_select_ntuple(
861 ep->com.dev->rdev.lldi.ports[0],
863 req6->opt2 = cpu_to_be32(opt2);
865 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
867 ep->com.dev->rdev.lldi.ports[0],
869 t5req6->rsvd = cpu_to_be32(isn);
870 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
871 t5req6->opt2 = cpu_to_be32(opt2);
875 set_bit(ACT_OPEN_REQ, &ep->com.history);
876 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
878 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
879 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
880 (const u32 *)&la6->sin6_addr.s6_addr, 1);
884 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
887 int mpalen, wrlen, ret;
888 struct fw_ofld_tx_data_wr *req;
889 struct mpa_message *mpa;
890 struct mpa_v2_conn_params mpa_v2_params;
892 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
894 BUG_ON(skb_cloned(skb));
896 mpalen = sizeof(*mpa) + ep->plen;
897 if (mpa_rev_to_use == 2)
898 mpalen += sizeof(struct mpa_v2_conn_params);
899 wrlen = roundup(mpalen + sizeof *req, 16);
900 skb = get_skb(skb, wrlen, GFP_KERNEL);
902 connect_reply_upcall(ep, -ENOMEM);
905 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
907 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
908 memset(req, 0, wrlen);
909 req->op_to_immdlen = cpu_to_be32(
910 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
912 FW_WR_IMMDLEN_V(mpalen));
913 req->flowid_len16 = cpu_to_be32(
914 FW_WR_FLOWID_V(ep->hwtid) |
915 FW_WR_LEN16_V(wrlen >> 4));
916 req->plen = cpu_to_be32(mpalen);
917 req->tunnel_to_proxy = cpu_to_be32(
918 FW_OFLD_TX_DATA_WR_FLUSH_F |
919 FW_OFLD_TX_DATA_WR_SHOVE_F);
921 mpa = (struct mpa_message *)(req + 1);
922 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
926 mpa->flags |= MPA_CRC;
927 if (markers_enabled) {
928 mpa->flags |= MPA_MARKERS;
929 ep->mpa_attr.recv_marker_enabled = 1;
931 ep->mpa_attr.recv_marker_enabled = 0;
933 if (mpa_rev_to_use == 2)
934 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
936 mpa->private_data_size = htons(ep->plen);
937 mpa->revision = mpa_rev_to_use;
938 if (mpa_rev_to_use == 1) {
939 ep->tried_with_mpa_v1 = 1;
940 ep->retry_with_mpa_v1 = 0;
943 if (mpa_rev_to_use == 2) {
944 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
945 sizeof (struct mpa_v2_conn_params));
946 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
948 mpa_v2_params.ird = htons((u16)ep->ird);
949 mpa_v2_params.ord = htons((u16)ep->ord);
952 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
953 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
955 htons(MPA_V2_RDMA_WRITE_RTR);
956 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
958 htons(MPA_V2_RDMA_READ_RTR);
960 memcpy(mpa->private_data, &mpa_v2_params,
961 sizeof(struct mpa_v2_conn_params));
964 memcpy(mpa->private_data +
965 sizeof(struct mpa_v2_conn_params),
966 ep->mpa_pkt + sizeof(*mpa), ep->plen);
969 memcpy(mpa->private_data,
970 ep->mpa_pkt + sizeof(*mpa), ep->plen);
973 * Reference the mpa skb. This ensures the data area
974 * will remain in memory until the hw acks the tx.
975 * Function fw4_ack() will deref it.
978 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
981 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
985 __state_set(&ep->com, MPA_REQ_SENT);
986 ep->mpa_attr.initiator = 1;
987 ep->snd_seq += mpalen;
991 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
994 struct fw_ofld_tx_data_wr *req;
995 struct mpa_message *mpa;
997 struct mpa_v2_conn_params mpa_v2_params;
999 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1001 mpalen = sizeof(*mpa) + plen;
1002 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1003 mpalen += sizeof(struct mpa_v2_conn_params);
1004 wrlen = roundup(mpalen + sizeof *req, 16);
1006 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1008 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1011 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1013 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1014 memset(req, 0, wrlen);
1015 req->op_to_immdlen = cpu_to_be32(
1016 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1018 FW_WR_IMMDLEN_V(mpalen));
1019 req->flowid_len16 = cpu_to_be32(
1020 FW_WR_FLOWID_V(ep->hwtid) |
1021 FW_WR_LEN16_V(wrlen >> 4));
1022 req->plen = cpu_to_be32(mpalen);
1023 req->tunnel_to_proxy = cpu_to_be32(
1024 FW_OFLD_TX_DATA_WR_FLUSH_F |
1025 FW_OFLD_TX_DATA_WR_SHOVE_F);
1027 mpa = (struct mpa_message *)(req + 1);
1028 memset(mpa, 0, sizeof(*mpa));
1029 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1030 mpa->flags = MPA_REJECT;
1031 mpa->revision = ep->mpa_attr.version;
1032 mpa->private_data_size = htons(plen);
1034 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1035 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1036 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1037 sizeof (struct mpa_v2_conn_params));
1038 mpa_v2_params.ird = htons(((u16)ep->ird) |
1039 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1041 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1043 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1044 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1045 FW_RI_INIT_P2PTYPE_READ_REQ ?
1046 MPA_V2_RDMA_READ_RTR : 0) : 0));
1047 memcpy(mpa->private_data, &mpa_v2_params,
1048 sizeof(struct mpa_v2_conn_params));
1051 memcpy(mpa->private_data +
1052 sizeof(struct mpa_v2_conn_params), pdata, plen);
1055 memcpy(mpa->private_data, pdata, plen);
1058 * Reference the mpa skb again. This ensures the data area
1059 * will remain in memory until the hw acks the tx.
1060 * Function fw4_ack() will deref it.
1063 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1064 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1065 BUG_ON(ep->mpa_skb);
1067 ep->snd_seq += mpalen;
1068 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1071 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1074 struct fw_ofld_tx_data_wr *req;
1075 struct mpa_message *mpa;
1076 struct sk_buff *skb;
1077 struct mpa_v2_conn_params mpa_v2_params;
1079 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1081 mpalen = sizeof(*mpa) + plen;
1082 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1083 mpalen += sizeof(struct mpa_v2_conn_params);
1084 wrlen = roundup(mpalen + sizeof *req, 16);
1086 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1088 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1091 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1093 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1094 memset(req, 0, wrlen);
1095 req->op_to_immdlen = cpu_to_be32(
1096 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1098 FW_WR_IMMDLEN_V(mpalen));
1099 req->flowid_len16 = cpu_to_be32(
1100 FW_WR_FLOWID_V(ep->hwtid) |
1101 FW_WR_LEN16_V(wrlen >> 4));
1102 req->plen = cpu_to_be32(mpalen);
1103 req->tunnel_to_proxy = cpu_to_be32(
1104 FW_OFLD_TX_DATA_WR_FLUSH_F |
1105 FW_OFLD_TX_DATA_WR_SHOVE_F);
1107 mpa = (struct mpa_message *)(req + 1);
1108 memset(mpa, 0, sizeof(*mpa));
1109 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1111 if (ep->mpa_attr.crc_enabled)
1112 mpa->flags |= MPA_CRC;
1113 if (ep->mpa_attr.recv_marker_enabled)
1114 mpa->flags |= MPA_MARKERS;
1115 mpa->revision = ep->mpa_attr.version;
1116 mpa->private_data_size = htons(plen);
1118 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1119 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1120 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1121 sizeof (struct mpa_v2_conn_params));
1122 mpa_v2_params.ird = htons((u16)ep->ird);
1123 mpa_v2_params.ord = htons((u16)ep->ord);
1124 if (peer2peer && (ep->mpa_attr.p2p_type !=
1125 FW_RI_INIT_P2PTYPE_DISABLED)) {
1126 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1128 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1129 mpa_v2_params.ord |=
1130 htons(MPA_V2_RDMA_WRITE_RTR);
1131 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1132 mpa_v2_params.ord |=
1133 htons(MPA_V2_RDMA_READ_RTR);
1136 memcpy(mpa->private_data, &mpa_v2_params,
1137 sizeof(struct mpa_v2_conn_params));
1140 memcpy(mpa->private_data +
1141 sizeof(struct mpa_v2_conn_params), pdata, plen);
1144 memcpy(mpa->private_data, pdata, plen);
1147 * Reference the mpa skb. This ensures the data area
1148 * will remain in memory until the hw acks the tx.
1149 * Function fw4_ack() will deref it.
1152 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1154 __state_set(&ep->com, MPA_REP_SENT);
1155 ep->snd_seq += mpalen;
1156 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1159 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1162 struct cpl_act_establish *req = cplhdr(skb);
1163 unsigned int tid = GET_TID(req);
1164 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1165 struct tid_info *t = dev->rdev.lldi.tids;
1168 ep = lookup_atid(t, atid);
1170 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1171 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1173 mutex_lock(&ep->com.mutex);
1174 dst_confirm(ep->dst);
1176 /* setup the hwtid for this connection */
1178 cxgb4_insert_tid(t, ep, tid);
1181 ep->snd_seq = be32_to_cpu(req->snd_isn);
1182 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1184 set_emss(ep, ntohs(req->tcp_opt));
1186 /* dealloc the atid */
1187 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1188 cxgb4_free_atid(t, atid);
1189 set_bit(ACT_ESTAB, &ep->com.history);
1191 /* start MPA negotiation */
1192 ret = send_flowc(ep);
1195 if (ep->retry_with_mpa_v1)
1196 ret = send_mpa_req(ep, skb, 1);
1198 ret = send_mpa_req(ep, skb, mpa_rev);
1201 mutex_unlock(&ep->com.mutex);
1204 mutex_unlock(&ep->com.mutex);
1205 connect_reply_upcall(ep, -ENOMEM);
1206 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1210 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1212 struct iw_cm_event event;
1214 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1215 memset(&event, 0, sizeof(event));
1216 event.event = IW_CM_EVENT_CLOSE;
1217 event.status = status;
1218 if (ep->com.cm_id) {
1219 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1220 ep, ep->com.cm_id, ep->hwtid);
1221 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1222 deref_cm_id(&ep->com);
1223 set_bit(CLOSE_UPCALL, &ep->com.history);
1227 static void peer_close_upcall(struct c4iw_ep *ep)
1229 struct iw_cm_event event;
1231 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1232 memset(&event, 0, sizeof(event));
1233 event.event = IW_CM_EVENT_DISCONNECT;
1234 if (ep->com.cm_id) {
1235 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1236 ep, ep->com.cm_id, ep->hwtid);
1237 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1238 set_bit(DISCONN_UPCALL, &ep->com.history);
1242 static void peer_abort_upcall(struct c4iw_ep *ep)
1244 struct iw_cm_event event;
1246 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1247 memset(&event, 0, sizeof(event));
1248 event.event = IW_CM_EVENT_CLOSE;
1249 event.status = -ECONNRESET;
1250 if (ep->com.cm_id) {
1251 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1252 ep->com.cm_id, ep->hwtid);
1253 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1254 deref_cm_id(&ep->com);
1255 set_bit(ABORT_UPCALL, &ep->com.history);
1259 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1261 struct iw_cm_event event;
1263 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1264 memset(&event, 0, sizeof(event));
1265 event.event = IW_CM_EVENT_CONNECT_REPLY;
1266 event.status = status;
1267 memcpy(&event.local_addr, &ep->com.local_addr,
1268 sizeof(ep->com.local_addr));
1269 memcpy(&event.remote_addr, &ep->com.remote_addr,
1270 sizeof(ep->com.remote_addr));
1272 if ((status == 0) || (status == -ECONNREFUSED)) {
1273 if (!ep->tried_with_mpa_v1) {
1274 /* this means MPA_v2 is used */
1275 event.ord = ep->ird;
1276 event.ird = ep->ord;
1277 event.private_data_len = ep->plen -
1278 sizeof(struct mpa_v2_conn_params);
1279 event.private_data = ep->mpa_pkt +
1280 sizeof(struct mpa_message) +
1281 sizeof(struct mpa_v2_conn_params);
1283 /* this means MPA_v1 is used */
1284 event.ord = cur_max_read_depth(ep->com.dev);
1285 event.ird = cur_max_read_depth(ep->com.dev);
1286 event.private_data_len = ep->plen;
1287 event.private_data = ep->mpa_pkt +
1288 sizeof(struct mpa_message);
1292 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1294 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1295 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1298 deref_cm_id(&ep->com);
1301 static int connect_request_upcall(struct c4iw_ep *ep)
1303 struct iw_cm_event event;
1306 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1307 memset(&event, 0, sizeof(event));
1308 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1309 memcpy(&event.local_addr, &ep->com.local_addr,
1310 sizeof(ep->com.local_addr));
1311 memcpy(&event.remote_addr, &ep->com.remote_addr,
1312 sizeof(ep->com.remote_addr));
1313 event.provider_data = ep;
1314 if (!ep->tried_with_mpa_v1) {
1315 /* this means MPA_v2 is used */
1316 event.ord = ep->ord;
1317 event.ird = ep->ird;
1318 event.private_data_len = ep->plen -
1319 sizeof(struct mpa_v2_conn_params);
1320 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1321 sizeof(struct mpa_v2_conn_params);
1323 /* this means MPA_v1 is used. Send max supported */
1324 event.ord = cur_max_read_depth(ep->com.dev);
1325 event.ird = cur_max_read_depth(ep->com.dev);
1326 event.private_data_len = ep->plen;
1327 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1329 c4iw_get_ep(&ep->com);
1330 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1333 c4iw_put_ep(&ep->com);
1334 set_bit(CONNREQ_UPCALL, &ep->com.history);
1335 c4iw_put_ep(&ep->parent_ep->com);
1339 static void established_upcall(struct c4iw_ep *ep)
1341 struct iw_cm_event event;
1343 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1344 memset(&event, 0, sizeof(event));
1345 event.event = IW_CM_EVENT_ESTABLISHED;
1346 event.ird = ep->ord;
1347 event.ord = ep->ird;
1348 if (ep->com.cm_id) {
1349 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1350 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1351 set_bit(ESTAB_UPCALL, &ep->com.history);
1355 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1357 struct sk_buff *skb;
1358 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1361 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1362 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1364 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1369 * If we couldn't specify the entire rcv window at connection setup
1370 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1371 * then add the overage in to the credits returned.
1373 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1374 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1376 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1377 RX_DACK_MODE_V(dack_mode);
1379 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1382 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1386 #define RELAXED_IRD_NEGOTIATION 1
1389 * process_mpa_reply - process streaming mode MPA reply
1393 * 0 upon success indicating a connect request was delivered to the ULP
1394 * or the mpa request is incomplete but valid so far.
1396 * 1 if a failure requires the caller to close the connection.
1398 * 2 if a failure requires the caller to abort the connection.
1400 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1402 struct mpa_message *mpa;
1403 struct mpa_v2_conn_params *mpa_v2_params;
1405 u16 resp_ird, resp_ord;
1406 u8 rtr_mismatch = 0, insuff_ird = 0;
1407 struct c4iw_qp_attributes attrs;
1408 enum c4iw_qp_attr_mask mask;
1412 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1415 * If we get more than the supported amount of private data
1416 * then we must fail this connection.
1418 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1420 goto err_stop_timer;
1424 * copy the new data into our accumulation buffer.
1426 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1428 ep->mpa_pkt_len += skb->len;
1431 * if we don't even have the mpa message, then bail.
1433 if (ep->mpa_pkt_len < sizeof(*mpa))
1435 mpa = (struct mpa_message *) ep->mpa_pkt;
1437 /* Validate MPA header. */
1438 if (mpa->revision > mpa_rev) {
1439 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1440 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1442 goto err_stop_timer;
1444 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1446 goto err_stop_timer;
1449 plen = ntohs(mpa->private_data_size);
1452 * Fail if there's too much private data.
1454 if (plen > MPA_MAX_PRIVATE_DATA) {
1456 goto err_stop_timer;
1460 * If plen does not account for pkt size
1462 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1464 goto err_stop_timer;
1467 ep->plen = (u8) plen;
1470 * If we don't have all the pdata yet, then bail.
1471 * We'll continue process when more data arrives.
1473 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1476 if (mpa->flags & MPA_REJECT) {
1477 err = -ECONNREFUSED;
1478 goto err_stop_timer;
1482 * Stop mpa timer. If it expired, then
1483 * we ignore the MPA reply. process_timeout()
1484 * will abort the connection.
1486 if (stop_ep_timer(ep))
1490 * If we get here we have accumulated the entire mpa
1491 * start reply message including private data. And
1492 * the MPA header is valid.
1494 __state_set(&ep->com, FPDU_MODE);
1495 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1496 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1497 ep->mpa_attr.version = mpa->revision;
1498 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1500 if (mpa->revision == 2) {
1501 ep->mpa_attr.enhanced_rdma_conn =
1502 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1503 if (ep->mpa_attr.enhanced_rdma_conn) {
1504 mpa_v2_params = (struct mpa_v2_conn_params *)
1505 (ep->mpa_pkt + sizeof(*mpa));
1506 resp_ird = ntohs(mpa_v2_params->ird) &
1507 MPA_V2_IRD_ORD_MASK;
1508 resp_ord = ntohs(mpa_v2_params->ord) &
1509 MPA_V2_IRD_ORD_MASK;
1510 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1511 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1514 * This is a double-check. Ideally, below checks are
1515 * not required since ird/ord stuff has been taken
1516 * care of in c4iw_accept_cr
1518 if (ep->ird < resp_ord) {
1519 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1520 ep->com.dev->rdev.lldi.max_ordird_qp)
1524 } else if (ep->ird > resp_ord) {
1527 if (ep->ord > resp_ird) {
1528 if (RELAXED_IRD_NEGOTIATION)
1539 if (ntohs(mpa_v2_params->ird) &
1540 MPA_V2_PEER2PEER_MODEL) {
1541 if (ntohs(mpa_v2_params->ord) &
1542 MPA_V2_RDMA_WRITE_RTR)
1543 ep->mpa_attr.p2p_type =
1544 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1545 else if (ntohs(mpa_v2_params->ord) &
1546 MPA_V2_RDMA_READ_RTR)
1547 ep->mpa_attr.p2p_type =
1548 FW_RI_INIT_P2PTYPE_READ_REQ;
1551 } else if (mpa->revision == 1)
1553 ep->mpa_attr.p2p_type = p2p_type;
1555 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1556 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1557 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1558 ep->mpa_attr.recv_marker_enabled,
1559 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1560 ep->mpa_attr.p2p_type, p2p_type);
1563 * If responder's RTR does not match with that of initiator, assign
1564 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1565 * generated when moving QP to RTS state.
1566 * A TERM message will be sent after QP has moved to RTS state
1568 if ((ep->mpa_attr.version == 2) && peer2peer &&
1569 (ep->mpa_attr.p2p_type != p2p_type)) {
1570 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1574 attrs.mpa_attr = ep->mpa_attr;
1575 attrs.max_ird = ep->ird;
1576 attrs.max_ord = ep->ord;
1577 attrs.llp_stream_handle = ep;
1578 attrs.next_state = C4IW_QP_STATE_RTS;
1580 mask = C4IW_QP_ATTR_NEXT_STATE |
1581 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1582 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1584 /* bind QP and TID with INIT_WR */
1585 err = c4iw_modify_qp(ep->com.qp->rhp,
1586 ep->com.qp, mask, &attrs, 1);
1591 * If responder's RTR requirement did not match with what initiator
1592 * supports, generate TERM message
1595 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1596 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1597 attrs.ecode = MPA_NOMATCH_RTR;
1598 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1599 attrs.send_term = 1;
1600 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1601 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1608 * Generate TERM if initiator IRD is not sufficient for responder
1609 * provided ORD. Currently, we do the same behaviour even when
1610 * responder provided IRD is also not sufficient as regards to
1614 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1616 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1617 attrs.ecode = MPA_INSUFF_IRD;
1618 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1619 attrs.send_term = 1;
1620 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1621 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1632 connect_reply_upcall(ep, err);
1637 * process_mpa_request - process streaming mode MPA request
1641 * 0 upon success indicating a connect request was delivered to the ULP
1642 * or the mpa request is incomplete but valid so far.
1644 * 1 if a failure requires the caller to close the connection.
1646 * 2 if a failure requires the caller to abort the connection.
1648 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1650 struct mpa_message *mpa;
1651 struct mpa_v2_conn_params *mpa_v2_params;
1654 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1657 * If we get more than the supported amount of private data
1658 * then we must fail this connection.
1660 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1661 goto err_stop_timer;
1663 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1666 * Copy the new data into our accumulation buffer.
1668 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1670 ep->mpa_pkt_len += skb->len;
1673 * If we don't even have the mpa message, then bail.
1674 * We'll continue process when more data arrives.
1676 if (ep->mpa_pkt_len < sizeof(*mpa))
1679 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1680 mpa = (struct mpa_message *) ep->mpa_pkt;
1683 * Validate MPA Header.
1685 if (mpa->revision > mpa_rev) {
1686 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1687 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1688 goto err_stop_timer;
1691 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1692 goto err_stop_timer;
1694 plen = ntohs(mpa->private_data_size);
1697 * Fail if there's too much private data.
1699 if (plen > MPA_MAX_PRIVATE_DATA)
1700 goto err_stop_timer;
1703 * If plen does not account for pkt size
1705 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1706 goto err_stop_timer;
1707 ep->plen = (u8) plen;
1710 * If we don't have all the pdata yet, then bail.
1712 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1716 * If we get here we have accumulated the entire mpa
1717 * start reply message including private data.
1719 ep->mpa_attr.initiator = 0;
1720 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1721 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1722 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1723 ep->mpa_attr.version = mpa->revision;
1724 if (mpa->revision == 1)
1725 ep->tried_with_mpa_v1 = 1;
1726 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1728 if (mpa->revision == 2) {
1729 ep->mpa_attr.enhanced_rdma_conn =
1730 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1731 if (ep->mpa_attr.enhanced_rdma_conn) {
1732 mpa_v2_params = (struct mpa_v2_conn_params *)
1733 (ep->mpa_pkt + sizeof(*mpa));
1734 ep->ird = ntohs(mpa_v2_params->ird) &
1735 MPA_V2_IRD_ORD_MASK;
1736 ep->ird = min_t(u32, ep->ird,
1737 cur_max_read_depth(ep->com.dev));
1738 ep->ord = ntohs(mpa_v2_params->ord) &
1739 MPA_V2_IRD_ORD_MASK;
1740 ep->ord = min_t(u32, ep->ord,
1741 cur_max_read_depth(ep->com.dev));
1742 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1744 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1746 if (ntohs(mpa_v2_params->ord) &
1747 MPA_V2_RDMA_WRITE_RTR)
1748 ep->mpa_attr.p2p_type =
1749 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1750 else if (ntohs(mpa_v2_params->ord) &
1751 MPA_V2_RDMA_READ_RTR)
1752 ep->mpa_attr.p2p_type =
1753 FW_RI_INIT_P2PTYPE_READ_REQ;
1756 } else if (mpa->revision == 1)
1758 ep->mpa_attr.p2p_type = p2p_type;
1760 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1761 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1762 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1763 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1764 ep->mpa_attr.p2p_type);
1766 __state_set(&ep->com, MPA_REQ_RCVD);
1769 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1770 if (ep->parent_ep->com.state != DEAD) {
1771 if (connect_request_upcall(ep))
1772 goto err_unlock_parent;
1774 goto err_unlock_parent;
1776 mutex_unlock(&ep->parent_ep->com.mutex);
1780 mutex_unlock(&ep->parent_ep->com.mutex);
1783 (void)stop_ep_timer(ep);
1788 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1791 struct cpl_rx_data *hdr = cplhdr(skb);
1792 unsigned int dlen = ntohs(hdr->len);
1793 unsigned int tid = GET_TID(hdr);
1794 __u8 status = hdr->status;
1797 ep = get_ep_from_tid(dev, tid);
1800 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1801 skb_pull(skb, sizeof(*hdr));
1802 skb_trim(skb, dlen);
1803 mutex_lock(&ep->com.mutex);
1805 /* update RX credits */
1806 update_rx_credits(ep, dlen);
1808 switch (ep->com.state) {
1810 ep->rcv_seq += dlen;
1811 disconnect = process_mpa_reply(ep, skb);
1814 ep->rcv_seq += dlen;
1815 disconnect = process_mpa_request(ep, skb);
1818 struct c4iw_qp_attributes attrs;
1819 BUG_ON(!ep->com.qp);
1821 pr_err("%s Unexpected streaming data." \
1822 " qpid %u ep %p state %d tid %u status %d\n",
1823 __func__, ep->com.qp->wq.sq.qid, ep,
1824 ep->com.state, ep->hwtid, status);
1825 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1826 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1827 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1834 mutex_unlock(&ep->com.mutex);
1836 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1837 c4iw_put_ep(&ep->com);
1841 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1844 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1846 unsigned int tid = GET_TID(rpl);
1848 ep = get_ep_from_tid(dev, tid);
1850 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1853 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1854 mutex_lock(&ep->com.mutex);
1855 switch (ep->com.state) {
1857 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1858 __state_set(&ep->com, DEAD);
1862 printk(KERN_ERR "%s ep %p state %d\n",
1863 __func__, ep, ep->com.state);
1866 mutex_unlock(&ep->com.mutex);
1869 release_ep_resources(ep);
1870 c4iw_put_ep(&ep->com);
1874 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1876 struct sk_buff *skb;
1877 struct fw_ofld_connection_wr *req;
1878 unsigned int mtu_idx;
1880 struct sockaddr_in *sin;
1883 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1884 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1885 memset(req, 0, sizeof(*req));
1886 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1887 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1888 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1889 ep->com.dev->rdev.lldi.ports[0],
1891 sin = (struct sockaddr_in *)&ep->com.local_addr;
1892 req->le.lport = sin->sin_port;
1893 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1894 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1895 req->le.pport = sin->sin_port;
1896 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1897 req->tcb.t_state_to_astid =
1898 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1899 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1900 req->tcb.cplrxdataack_cplpassacceptrpl =
1901 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1902 req->tcb.tx_max = (__force __be32) jiffies;
1903 req->tcb.rcv_adv = htons(1);
1904 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1905 enable_tcp_timestamps,
1906 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1907 wscale = cxgb_compute_wscale(rcv_win);
1910 * Specify the largest window that will fit in opt0. The
1911 * remainder will be specified in the rx_data_ack.
1913 win = ep->rcv_win >> 10;
1914 if (win > RCV_BUFSIZ_M)
1917 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1918 (nocong ? NO_CONG_F : 0) |
1921 WND_SCALE_V(wscale) |
1922 MSS_IDX_V(mtu_idx) |
1923 L2T_IDX_V(ep->l2t->idx) |
1924 TX_CHAN_V(ep->tx_chan) |
1925 SMAC_SEL_V(ep->smac_idx) |
1926 DSCP_V(ep->tos >> 2) |
1927 ULP_MODE_V(ULP_MODE_TCPDDP) |
1929 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1930 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1932 CCTRL_ECN_V(enable_ecn) |
1933 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1934 if (enable_tcp_timestamps)
1935 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1936 if (enable_tcp_sack)
1937 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1938 if (wscale && enable_tcp_window_scaling)
1939 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1940 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1941 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1942 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1943 set_bit(ACT_OFLD_CONN, &ep->com.history);
1944 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1948 * Some of the error codes above implicitly indicate that there is no TID
1949 * allocated with the result of an ACT_OPEN. We use this predicate to make
1952 static inline int act_open_has_tid(int status)
1954 return (status != CPL_ERR_TCAM_PARITY &&
1955 status != CPL_ERR_TCAM_MISS &&
1956 status != CPL_ERR_TCAM_FULL &&
1957 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1958 status != CPL_ERR_CONN_EXIST);
1961 static char *neg_adv_str(unsigned int status)
1964 case CPL_ERR_RTX_NEG_ADVICE:
1965 return "Retransmit timeout";
1966 case CPL_ERR_PERSIST_NEG_ADVICE:
1967 return "Persist timeout";
1968 case CPL_ERR_KEEPALV_NEG_ADVICE:
1969 return "Keepalive timeout";
1975 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1977 ep->snd_win = snd_win;
1978 ep->rcv_win = rcv_win;
1979 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
1982 #define ACT_OPEN_RETRY_COUNT 2
1984 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1985 struct dst_entry *dst, struct c4iw_dev *cdev,
1986 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
1988 struct neighbour *n;
1990 struct net_device *pdev;
1992 n = dst_neigh_lookup(dst, peer_ip);
1998 if (n->dev->flags & IFF_LOOPBACK) {
2000 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2001 else if (IS_ENABLED(CONFIG_IPV6))
2002 for_each_netdev(&init_net, pdev) {
2003 if (ipv6_chk_addr(&init_net,
2004 (struct in6_addr *)peer_ip,
2015 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2016 n, pdev, rt_tos2priority(tos));
2019 ep->mtu = pdev->mtu;
2020 ep->tx_chan = cxgb4_port_chan(pdev);
2021 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2022 cxgb4_port_viid(pdev));
2023 step = cdev->rdev.lldi.ntxq /
2024 cdev->rdev.lldi.nchan;
2025 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2026 step = cdev->rdev.lldi.nrxq /
2027 cdev->rdev.lldi.nchan;
2028 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2029 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2030 cxgb4_port_idx(pdev) * step];
2031 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2034 pdev = get_real_dev(n->dev);
2035 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2039 ep->mtu = dst_mtu(dst);
2040 ep->tx_chan = cxgb4_port_chan(pdev);
2041 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2042 cxgb4_port_viid(pdev));
2043 step = cdev->rdev.lldi.ntxq /
2044 cdev->rdev.lldi.nchan;
2045 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2046 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2047 step = cdev->rdev.lldi.nrxq /
2048 cdev->rdev.lldi.nchan;
2049 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2050 cxgb4_port_idx(pdev) * step];
2051 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2054 ep->retry_with_mpa_v1 = 0;
2055 ep->tried_with_mpa_v1 = 0;
2067 static int c4iw_reconnect(struct c4iw_ep *ep)
2071 struct sockaddr_in *laddr = (struct sockaddr_in *)
2072 &ep->com.cm_id->m_local_addr;
2073 struct sockaddr_in *raddr = (struct sockaddr_in *)
2074 &ep->com.cm_id->m_remote_addr;
2075 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2076 &ep->com.cm_id->m_local_addr;
2077 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2078 &ep->com.cm_id->m_remote_addr;
2082 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2083 init_timer(&ep->timer);
2084 c4iw_init_wr_wait(&ep->com.wr_wait);
2086 /* When MPA revision is different on nodes, the node with MPA_rev=2
2087 * tries to reconnect with MPA_rev 1 for the same EP through
2088 * c4iw_reconnect(), where the same EP is assigned with new tid for
2089 * further connection establishment. As we are using the same EP pointer
2090 * for reconnect, few skbs are used during the previous c4iw_connect(),
2091 * which leaves the EP with inadequate skbs for further
2092 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2093 * skb_list() during peer_abort(). Allocate skbs which is already used.
2095 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2096 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2102 * Allocate an active TID to initiate a TCP connection.
2104 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2105 if (ep->atid == -1) {
2106 pr_err("%s - cannot alloc atid.\n", __func__);
2110 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2113 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2114 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2115 laddr->sin_addr.s_addr,
2116 raddr->sin_addr.s_addr,
2118 raddr->sin_port, ep->com.cm_id->tos);
2120 ra = (__u8 *)&raddr->sin_addr;
2122 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2124 laddr6->sin6_addr.s6_addr,
2125 raddr6->sin6_addr.s6_addr,
2127 raddr6->sin6_port, 0,
2128 raddr6->sin6_scope_id);
2130 ra = (__u8 *)&raddr6->sin6_addr;
2133 pr_err("%s - cannot find route.\n", __func__);
2134 err = -EHOSTUNREACH;
2137 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2138 ep->com.dev->rdev.lldi.adapter_type,
2139 ep->com.cm_id->tos);
2141 pr_err("%s - cannot alloc l2e.\n", __func__);
2145 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2146 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2149 state_set(&ep->com, CONNECTING);
2150 ep->tos = ep->com.cm_id->tos;
2152 /* send connect request to rnic */
2153 err = send_connect(ep);
2157 cxgb4_l2t_release(ep->l2t);
2159 dst_release(ep->dst);
2161 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2162 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2165 * remember to send notification to upper layer.
2166 * We are in here so the upper layer is not aware that this is
2167 * re-connect attempt and so, upper layer is still waiting for
2168 * response of 1st connect request.
2170 connect_reply_upcall(ep, -ECONNRESET);
2172 c4iw_put_ep(&ep->com);
2177 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2180 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2181 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2182 ntohl(rpl->atid_status)));
2183 struct tid_info *t = dev->rdev.lldi.tids;
2184 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2185 struct sockaddr_in *la;
2186 struct sockaddr_in *ra;
2187 struct sockaddr_in6 *la6;
2188 struct sockaddr_in6 *ra6;
2191 ep = lookup_atid(t, atid);
2192 la = (struct sockaddr_in *)&ep->com.local_addr;
2193 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2194 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2195 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2197 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2198 status, status2errno(status));
2200 if (cxgb_is_neg_adv(status)) {
2201 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2202 __func__, atid, status, neg_adv_str(status));
2203 ep->stats.connect_neg_adv++;
2204 mutex_lock(&dev->rdev.stats.lock);
2205 dev->rdev.stats.neg_adv++;
2206 mutex_unlock(&dev->rdev.stats.lock);
2210 set_bit(ACT_OPEN_RPL, &ep->com.history);
2213 * Log interesting failures.
2216 case CPL_ERR_CONN_RESET:
2217 case CPL_ERR_CONN_TIMEDOUT:
2219 case CPL_ERR_TCAM_FULL:
2220 mutex_lock(&dev->rdev.stats.lock);
2221 dev->rdev.stats.tcam_full++;
2222 mutex_unlock(&dev->rdev.stats.lock);
2223 if (ep->com.local_addr.ss_family == AF_INET &&
2224 dev->rdev.lldi.enable_fw_ofld_conn) {
2225 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2226 ntohl(rpl->atid_status))));
2232 case CPL_ERR_CONN_EXIST:
2233 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2234 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2235 if (ep->com.remote_addr.ss_family == AF_INET6) {
2236 struct sockaddr_in6 *sin6 =
2237 (struct sockaddr_in6 *)
2238 &ep->com.local_addr;
2240 ep->com.dev->rdev.lldi.ports[0],
2242 &sin6->sin6_addr.s6_addr, 1);
2244 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2246 cxgb4_free_atid(t, atid);
2247 dst_release(ep->dst);
2248 cxgb4_l2t_release(ep->l2t);
2254 if (ep->com.local_addr.ss_family == AF_INET) {
2255 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2256 atid, status, status2errno(status),
2257 &la->sin_addr.s_addr, ntohs(la->sin_port),
2258 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2260 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2261 atid, status, status2errno(status),
2262 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2263 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2269 connect_reply_upcall(ep, status2errno(status));
2270 state_set(&ep->com, DEAD);
2272 if (ep->com.remote_addr.ss_family == AF_INET6) {
2273 struct sockaddr_in6 *sin6 =
2274 (struct sockaddr_in6 *)&ep->com.local_addr;
2275 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2276 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2278 if (status && act_open_has_tid(status))
2279 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2281 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2282 cxgb4_free_atid(t, atid);
2283 dst_release(ep->dst);
2284 cxgb4_l2t_release(ep->l2t);
2285 c4iw_put_ep(&ep->com);
2290 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2292 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2293 unsigned int stid = GET_TID(rpl);
2294 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2297 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2300 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2301 rpl->status, status2errno(rpl->status));
2302 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2303 c4iw_put_ep(&ep->com);
2308 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2310 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2311 unsigned int stid = GET_TID(rpl);
2312 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2314 PDBG("%s ep %p\n", __func__, ep);
2315 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2316 c4iw_put_ep(&ep->com);
2320 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2321 struct cpl_pass_accept_req *req)
2323 struct cpl_pass_accept_rpl *rpl;
2324 unsigned int mtu_idx;
2328 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2330 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2332 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2333 BUG_ON(skb_cloned(skb));
2337 if (!is_t4(adapter_type)) {
2338 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2340 INIT_TP_WR(rpl5, ep->hwtid);
2342 skb_trim(skb, sizeof(*rpl));
2343 INIT_TP_WR(rpl, ep->hwtid);
2345 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2348 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2349 enable_tcp_timestamps && req->tcpopt.tstamp,
2350 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2351 wscale = cxgb_compute_wscale(rcv_win);
2354 * Specify the largest window that will fit in opt0. The
2355 * remainder will be specified in the rx_data_ack.
2357 win = ep->rcv_win >> 10;
2358 if (win > RCV_BUFSIZ_M)
2360 opt0 = (nocong ? NO_CONG_F : 0) |
2363 WND_SCALE_V(wscale) |
2364 MSS_IDX_V(mtu_idx) |
2365 L2T_IDX_V(ep->l2t->idx) |
2366 TX_CHAN_V(ep->tx_chan) |
2367 SMAC_SEL_V(ep->smac_idx) |
2368 DSCP_V(ep->tos >> 2) |
2369 ULP_MODE_V(ULP_MODE_TCPDDP) |
2371 opt2 = RX_CHANNEL_V(0) |
2372 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2374 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2375 opt2 |= TSTAMPS_EN_F;
2376 if (enable_tcp_sack && req->tcpopt.sack)
2378 if (wscale && enable_tcp_window_scaling)
2379 opt2 |= WND_SCALE_EN_F;
2381 const struct tcphdr *tcph;
2382 u32 hlen = ntohl(req->hdr_len);
2384 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2385 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2388 tcph = (const void *)(req + 1) +
2389 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2390 if (tcph->ece && tcph->cwr)
2391 opt2 |= CCTRL_ECN_V(1);
2393 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2394 u32 isn = (prandom_u32() & ~7UL) - 1;
2395 opt2 |= T5_OPT_2_VALID_F;
2396 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2399 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2402 rpl5->iss = cpu_to_be32(isn);
2403 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2406 rpl->opt0 = cpu_to_be64(opt0);
2407 rpl->opt2 = cpu_to_be32(opt2);
2408 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2409 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2411 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2414 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2416 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2417 BUG_ON(skb_cloned(skb));
2418 skb_trim(skb, sizeof(struct cpl_tid_release));
2419 release_tid(&dev->rdev, hwtid, skb);
2423 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2425 struct c4iw_ep *child_ep = NULL, *parent_ep;
2426 struct cpl_pass_accept_req *req = cplhdr(skb);
2427 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2428 struct tid_info *t = dev->rdev.lldi.tids;
2429 unsigned int hwtid = GET_TID(req);
2430 struct dst_entry *dst;
2431 __u8 local_ip[16], peer_ip[16];
2432 __be16 local_port, peer_port;
2433 struct sockaddr_in6 *sin6;
2435 u16 peer_mss = ntohs(req->tcpopt.mss);
2437 unsigned short hdrs;
2438 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2440 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2442 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2446 if (state_read(&parent_ep->com) != LISTEN) {
2447 PDBG("%s - listening ep not in LISTEN\n", __func__);
2451 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2452 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2454 /* Find output route */
2456 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2457 , __func__, parent_ep, hwtid,
2458 local_ip, peer_ip, ntohs(local_port),
2459 ntohs(peer_port), peer_mss);
2460 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2461 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2462 local_port, peer_port, tos);
2464 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2465 , __func__, parent_ep, hwtid,
2466 local_ip, peer_ip, ntohs(local_port),
2467 ntohs(peer_port), peer_mss);
2468 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2469 local_ip, peer_ip, local_port, peer_port,
2470 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2471 ((struct sockaddr_in6 *)
2472 &parent_ep->com.local_addr)->sin6_scope_id);
2475 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2480 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2482 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2488 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2489 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2491 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2498 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2499 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2500 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2501 child_ep->mtu = peer_mss + hdrs;
2503 skb_queue_head_init(&child_ep->com.ep_skb_list);
2504 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2507 state_set(&child_ep->com, CONNECTING);
2508 child_ep->com.dev = dev;
2509 child_ep->com.cm_id = NULL;
2512 struct sockaddr_in *sin = (struct sockaddr_in *)
2513 &child_ep->com.local_addr;
2515 sin->sin_family = PF_INET;
2516 sin->sin_port = local_port;
2517 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2519 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2520 sin->sin_family = PF_INET;
2521 sin->sin_port = ((struct sockaddr_in *)
2522 &parent_ep->com.local_addr)->sin_port;
2523 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2525 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2526 sin->sin_family = PF_INET;
2527 sin->sin_port = peer_port;
2528 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2530 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2531 sin6->sin6_family = PF_INET6;
2532 sin6->sin6_port = local_port;
2533 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2535 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2536 sin6->sin6_family = PF_INET6;
2537 sin6->sin6_port = ((struct sockaddr_in6 *)
2538 &parent_ep->com.local_addr)->sin6_port;
2539 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2541 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2542 sin6->sin6_family = PF_INET6;
2543 sin6->sin6_port = peer_port;
2544 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2547 c4iw_get_ep(&parent_ep->com);
2548 child_ep->parent_ep = parent_ep;
2549 child_ep->tos = tos;
2550 child_ep->dst = dst;
2551 child_ep->hwtid = hwtid;
2553 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2554 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2556 init_timer(&child_ep->timer);
2557 cxgb4_insert_tid(t, child_ep, hwtid);
2558 insert_ep_tid(child_ep);
2559 if (accept_cr(child_ep, skb, req)) {
2560 c4iw_put_ep(&parent_ep->com);
2561 release_ep_resources(child_ep);
2563 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2566 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2567 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2568 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2572 c4iw_put_ep(&child_ep->com);
2574 reject_cr(dev, hwtid, skb);
2576 c4iw_put_ep(&parent_ep->com);
2581 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2584 struct cpl_pass_establish *req = cplhdr(skb);
2585 unsigned int tid = GET_TID(req);
2588 ep = get_ep_from_tid(dev, tid);
2589 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2590 ep->snd_seq = be32_to_cpu(req->snd_isn);
2591 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2593 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2594 ntohs(req->tcp_opt));
2596 set_emss(ep, ntohs(req->tcp_opt));
2598 dst_confirm(ep->dst);
2599 mutex_lock(&ep->com.mutex);
2600 ep->com.state = MPA_REQ_WAIT;
2602 set_bit(PASS_ESTAB, &ep->com.history);
2603 ret = send_flowc(ep);
2604 mutex_unlock(&ep->com.mutex);
2606 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2607 c4iw_put_ep(&ep->com);
2612 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2614 struct cpl_peer_close *hdr = cplhdr(skb);
2616 struct c4iw_qp_attributes attrs;
2619 unsigned int tid = GET_TID(hdr);
2622 ep = get_ep_from_tid(dev, tid);
2626 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2627 dst_confirm(ep->dst);
2629 set_bit(PEER_CLOSE, &ep->com.history);
2630 mutex_lock(&ep->com.mutex);
2631 switch (ep->com.state) {
2633 __state_set(&ep->com, CLOSING);
2636 __state_set(&ep->com, CLOSING);
2637 connect_reply_upcall(ep, -ECONNRESET);
2642 * We're gonna mark this puppy DEAD, but keep
2643 * the reference on it until the ULP accepts or
2644 * rejects the CR. Also wake up anyone waiting
2645 * in rdma connection migration (see c4iw_accept_cr()).
2647 __state_set(&ep->com, CLOSING);
2648 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2649 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2652 __state_set(&ep->com, CLOSING);
2653 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2654 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2658 __state_set(&ep->com, CLOSING);
2659 attrs.next_state = C4IW_QP_STATE_CLOSING;
2660 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2661 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2662 if (ret != -ECONNRESET) {
2663 peer_close_upcall(ep);
2671 __state_set(&ep->com, MORIBUND);
2675 (void)stop_ep_timer(ep);
2676 if (ep->com.cm_id && ep->com.qp) {
2677 attrs.next_state = C4IW_QP_STATE_IDLE;
2678 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2679 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2681 close_complete_upcall(ep, 0);
2682 __state_set(&ep->com, DEAD);
2692 mutex_unlock(&ep->com.mutex);
2694 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2696 release_ep_resources(ep);
2697 c4iw_put_ep(&ep->com);
2701 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2703 struct cpl_abort_req_rss *req = cplhdr(skb);
2705 struct sk_buff *rpl_skb;
2706 struct c4iw_qp_attributes attrs;
2709 unsigned int tid = GET_TID(req);
2710 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2712 ep = get_ep_from_tid(dev, tid);
2716 if (cxgb_is_neg_adv(req->status)) {
2717 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2718 __func__, ep->hwtid, req->status,
2719 neg_adv_str(req->status));
2720 ep->stats.abort_neg_adv++;
2721 mutex_lock(&dev->rdev.stats.lock);
2722 dev->rdev.stats.neg_adv++;
2723 mutex_unlock(&dev->rdev.stats.lock);
2726 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2728 set_bit(PEER_ABORT, &ep->com.history);
2731 * Wake up any threads in rdma_init() or rdma_fini().
2732 * However, this is not needed if com state is just
2735 if (ep->com.state != MPA_REQ_SENT)
2736 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2738 mutex_lock(&ep->com.mutex);
2739 switch (ep->com.state) {
2741 c4iw_put_ep(&ep->parent_ep->com);
2744 (void)stop_ep_timer(ep);
2747 (void)stop_ep_timer(ep);
2748 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2749 connect_reply_upcall(ep, -ECONNRESET);
2752 * we just don't send notification upwards because we
2753 * want to retry with mpa_v1 without upper layers even
2756 * do some housekeeping so as to re-initiate the
2759 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2761 ep->retry_with_mpa_v1 = 1;
2773 if (ep->com.cm_id && ep->com.qp) {
2774 attrs.next_state = C4IW_QP_STATE_ERROR;
2775 ret = c4iw_modify_qp(ep->com.qp->rhp,
2776 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2780 "%s - qp <- error failed!\n",
2783 peer_abort_upcall(ep);
2788 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2789 mutex_unlock(&ep->com.mutex);
2795 dst_confirm(ep->dst);
2796 if (ep->com.state != ABORTING) {
2797 __state_set(&ep->com, DEAD);
2798 /* we don't release if we want to retry with mpa_v1 */
2799 if (!ep->retry_with_mpa_v1)
2802 mutex_unlock(&ep->com.mutex);
2804 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2805 if (WARN_ON(!rpl_skb)) {
2810 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2812 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2815 release_ep_resources(ep);
2816 else if (ep->retry_with_mpa_v1) {
2817 if (ep->com.remote_addr.ss_family == AF_INET6) {
2818 struct sockaddr_in6 *sin6 =
2819 (struct sockaddr_in6 *)
2820 &ep->com.local_addr;
2822 ep->com.dev->rdev.lldi.ports[0],
2823 (const u32 *)&sin6->sin6_addr.s6_addr,
2826 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2827 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2828 dst_release(ep->dst);
2829 cxgb4_l2t_release(ep->l2t);
2834 c4iw_put_ep(&ep->com);
2835 /* Dereferencing ep, referenced in peer_abort_intr() */
2836 c4iw_put_ep(&ep->com);
2840 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2843 struct c4iw_qp_attributes attrs;
2844 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2846 unsigned int tid = GET_TID(rpl);
2848 ep = get_ep_from_tid(dev, tid);
2852 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2855 /* The cm_id may be null if we failed to connect */
2856 mutex_lock(&ep->com.mutex);
2857 set_bit(CLOSE_CON_RPL, &ep->com.history);
2858 switch (ep->com.state) {
2860 __state_set(&ep->com, MORIBUND);
2863 (void)stop_ep_timer(ep);
2864 if ((ep->com.cm_id) && (ep->com.qp)) {
2865 attrs.next_state = C4IW_QP_STATE_IDLE;
2866 c4iw_modify_qp(ep->com.qp->rhp,
2868 C4IW_QP_ATTR_NEXT_STATE,
2871 close_complete_upcall(ep, 0);
2872 __state_set(&ep->com, DEAD);
2882 mutex_unlock(&ep->com.mutex);
2884 release_ep_resources(ep);
2885 c4iw_put_ep(&ep->com);
2889 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2891 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2892 unsigned int tid = GET_TID(rpl);
2894 struct c4iw_qp_attributes attrs;
2896 ep = get_ep_from_tid(dev, tid);
2899 if (ep && ep->com.qp) {
2900 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2901 ep->com.qp->wq.sq.qid);
2902 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2903 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2904 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2906 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2907 c4iw_put_ep(&ep->com);
2913 * Upcall from the adapter indicating data has been transmitted.
2914 * For us its just the single MPA request or reply. We can now free
2915 * the skb holding the mpa message.
2917 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2920 struct cpl_fw4_ack *hdr = cplhdr(skb);
2921 u8 credits = hdr->credits;
2922 unsigned int tid = GET_TID(hdr);
2925 ep = get_ep_from_tid(dev, tid);
2928 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2930 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2931 __func__, ep, ep->hwtid, state_read(&ep->com));
2935 dst_confirm(ep->dst);
2937 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2938 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2939 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2940 mutex_lock(&ep->com.mutex);
2941 kfree_skb(ep->mpa_skb);
2943 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2945 mutex_unlock(&ep->com.mutex);
2948 c4iw_put_ep(&ep->com);
2952 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2955 struct c4iw_ep *ep = to_ep(cm_id);
2957 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2959 mutex_lock(&ep->com.mutex);
2960 if (ep->com.state != MPA_REQ_RCVD) {
2961 mutex_unlock(&ep->com.mutex);
2962 c4iw_put_ep(&ep->com);
2965 set_bit(ULP_REJECT, &ep->com.history);
2969 abort = send_mpa_reject(ep, pdata, pdata_len);
2970 mutex_unlock(&ep->com.mutex);
2973 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
2974 c4iw_put_ep(&ep->com);
2978 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2981 struct c4iw_qp_attributes attrs;
2982 enum c4iw_qp_attr_mask mask;
2983 struct c4iw_ep *ep = to_ep(cm_id);
2984 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2985 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2988 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2990 mutex_lock(&ep->com.mutex);
2991 if (ep->com.state != MPA_REQ_RCVD) {
2998 set_bit(ULP_ACCEPT, &ep->com.history);
2999 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3000 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3005 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3006 if (conn_param->ord > ep->ird) {
3007 if (RELAXED_IRD_NEGOTIATION) {
3008 conn_param->ord = ep->ird;
3010 ep->ird = conn_param->ird;
3011 ep->ord = conn_param->ord;
3012 send_mpa_reject(ep, conn_param->private_data,
3013 conn_param->private_data_len);
3018 if (conn_param->ird < ep->ord) {
3019 if (RELAXED_IRD_NEGOTIATION &&
3020 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3021 conn_param->ird = ep->ord;
3028 ep->ird = conn_param->ird;
3029 ep->ord = conn_param->ord;
3031 if (ep->mpa_attr.version == 1) {
3032 if (peer2peer && ep->ird == 0)
3036 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3037 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3041 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3043 ep->com.cm_id = cm_id;
3044 ref_cm_id(&ep->com);
3048 /* bind QP to EP and move to RTS */
3049 attrs.mpa_attr = ep->mpa_attr;
3050 attrs.max_ird = ep->ird;
3051 attrs.max_ord = ep->ord;
3052 attrs.llp_stream_handle = ep;
3053 attrs.next_state = C4IW_QP_STATE_RTS;
3055 /* bind QP and TID with INIT_WR */
3056 mask = C4IW_QP_ATTR_NEXT_STATE |
3057 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3058 C4IW_QP_ATTR_MPA_ATTR |
3059 C4IW_QP_ATTR_MAX_IRD |
3060 C4IW_QP_ATTR_MAX_ORD;
3062 err = c4iw_modify_qp(ep->com.qp->rhp,
3063 ep->com.qp, mask, &attrs, 1);
3065 goto err_deref_cm_id;
3067 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3068 err = send_mpa_reply(ep, conn_param->private_data,
3069 conn_param->private_data_len);
3071 goto err_deref_cm_id;
3073 __state_set(&ep->com, FPDU_MODE);
3074 established_upcall(ep);
3075 mutex_unlock(&ep->com.mutex);
3076 c4iw_put_ep(&ep->com);
3079 deref_cm_id(&ep->com);
3083 mutex_unlock(&ep->com.mutex);
3085 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3086 c4iw_put_ep(&ep->com);
3090 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3092 struct in_device *ind;
3094 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3095 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3097 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3099 return -EADDRNOTAVAIL;
3100 for_primary_ifa(ind) {
3101 laddr->sin_addr.s_addr = ifa->ifa_address;
3102 raddr->sin_addr.s_addr = ifa->ifa_address;
3108 return found ? 0 : -EADDRNOTAVAIL;
3111 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3112 unsigned char banned_flags)
3114 struct inet6_dev *idev;
3115 int err = -EADDRNOTAVAIL;
3118 idev = __in6_dev_get(dev);
3120 struct inet6_ifaddr *ifp;
3122 read_lock_bh(&idev->lock);
3123 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3124 if (ifp->scope == IFA_LINK &&
3125 !(ifp->flags & banned_flags)) {
3126 memcpy(addr, &ifp->addr, 16);
3131 read_unlock_bh(&idev->lock);
3137 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3139 struct in6_addr uninitialized_var(addr);
3140 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3141 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3143 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3144 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3145 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3148 return -EADDRNOTAVAIL;
3151 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3153 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3156 struct sockaddr_in *laddr;
3157 struct sockaddr_in *raddr;
3158 struct sockaddr_in6 *laddr6;
3159 struct sockaddr_in6 *raddr6;
3163 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3164 (conn_param->ird > cur_max_read_depth(dev))) {
3168 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3170 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3175 skb_queue_head_init(&ep->com.ep_skb_list);
3176 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3181 init_timer(&ep->timer);
3182 ep->plen = conn_param->private_data_len;
3184 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3185 conn_param->private_data, ep->plen);
3186 ep->ird = conn_param->ird;
3187 ep->ord = conn_param->ord;
3189 if (peer2peer && ep->ord == 0)
3192 ep->com.cm_id = cm_id;
3193 ref_cm_id(&ep->com);
3195 ep->com.qp = get_qhp(dev, conn_param->qpn);
3197 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3202 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3206 * Allocate an active TID to initiate a TCP connection.
3208 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3209 if (ep->atid == -1) {
3210 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3214 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3216 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3217 sizeof(ep->com.local_addr));
3218 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3219 sizeof(ep->com.remote_addr));
3221 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3222 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3223 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3224 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3226 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3228 ra = (__u8 *)&raddr->sin_addr;
3231 * Handle loopback requests to INADDR_ANY.
3233 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3234 err = pick_local_ipaddrs(dev, cm_id);
3240 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3241 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3242 ra, ntohs(raddr->sin_port));
3243 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3244 laddr->sin_addr.s_addr,
3245 raddr->sin_addr.s_addr,
3247 raddr->sin_port, cm_id->tos);
3250 ra = (__u8 *)&raddr6->sin6_addr;
3253 * Handle loopback requests to INADDR_ANY.
3255 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3256 err = pick_local_ip6addrs(dev, cm_id);
3262 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3263 __func__, laddr6->sin6_addr.s6_addr,
3264 ntohs(laddr6->sin6_port),
3265 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3266 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3267 laddr6->sin6_addr.s6_addr,
3268 raddr6->sin6_addr.s6_addr,
3270 raddr6->sin6_port, 0,
3271 raddr6->sin6_scope_id);
3274 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3275 err = -EHOSTUNREACH;
3279 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3280 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3282 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3286 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3287 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3290 state_set(&ep->com, CONNECTING);
3291 ep->tos = cm_id->tos;
3293 /* send connect request to rnic */
3294 err = send_connect(ep);
3298 cxgb4_l2t_release(ep->l2t);
3300 dst_release(ep->dst);
3302 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3303 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3305 skb_queue_purge(&ep->com.ep_skb_list);
3306 deref_cm_id(&ep->com);
3308 c4iw_put_ep(&ep->com);
3313 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3316 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3317 &ep->com.local_addr;
3319 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3320 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3321 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3325 c4iw_init_wr_wait(&ep->com.wr_wait);
3326 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3327 ep->stid, &sin6->sin6_addr,
3329 ep->com.dev->rdev.lldi.rxq_ids[0]);
3331 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3335 err = net_xmit_errno(err);
3337 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3338 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3339 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3341 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3346 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3349 struct sockaddr_in *sin = (struct sockaddr_in *)
3350 &ep->com.local_addr;
3352 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3354 err = cxgb4_create_server_filter(
3355 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3356 sin->sin_addr.s_addr, sin->sin_port, 0,
3357 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3358 if (err == -EBUSY) {
3359 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3363 set_current_state(TASK_UNINTERRUPTIBLE);
3364 schedule_timeout(usecs_to_jiffies(100));
3366 } while (err == -EBUSY);
3368 c4iw_init_wr_wait(&ep->com.wr_wait);
3369 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3370 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3371 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3373 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3377 err = net_xmit_errno(err);
3380 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3382 &sin->sin_addr, ntohs(sin->sin_port));
3386 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3389 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3390 struct c4iw_listen_ep *ep;
3394 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3396 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3400 skb_queue_head_init(&ep->com.ep_skb_list);
3401 PDBG("%s ep %p\n", __func__, ep);
3402 ep->com.cm_id = cm_id;
3403 ref_cm_id(&ep->com);
3405 ep->backlog = backlog;
3406 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3407 sizeof(ep->com.local_addr));
3410 * Allocate a server TID.
3412 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3413 ep->com.local_addr.ss_family == AF_INET)
3414 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3415 cm_id->m_local_addr.ss_family, ep);
3417 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3418 cm_id->m_local_addr.ss_family, ep);
3420 if (ep->stid == -1) {
3421 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3425 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3427 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3428 sizeof(ep->com.local_addr));
3430 state_set(&ep->com, LISTEN);
3431 if (ep->com.local_addr.ss_family == AF_INET)
3432 err = create_server4(dev, ep);
3434 err = create_server6(dev, ep);
3436 cm_id->provider_data = ep;
3440 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3441 ep->com.local_addr.ss_family);
3443 deref_cm_id(&ep->com);
3444 c4iw_put_ep(&ep->com);
3450 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3453 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3455 PDBG("%s ep %p\n", __func__, ep);
3458 state_set(&ep->com, DEAD);
3459 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3460 ep->com.local_addr.ss_family == AF_INET) {
3461 err = cxgb4_remove_server_filter(
3462 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3463 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3465 struct sockaddr_in6 *sin6;
3466 c4iw_init_wr_wait(&ep->com.wr_wait);
3467 err = cxgb4_remove_server(
3468 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3469 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3472 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3474 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3475 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3476 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3478 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3479 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3480 ep->com.local_addr.ss_family);
3482 deref_cm_id(&ep->com);
3483 c4iw_put_ep(&ep->com);
3487 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3492 struct c4iw_rdev *rdev;
3494 mutex_lock(&ep->com.mutex);
3496 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3497 states[ep->com.state], abrupt);
3500 * Ref the ep here in case we have fatal errors causing the
3501 * ep to be released and freed.
3503 c4iw_get_ep(&ep->com);
3505 rdev = &ep->com.dev->rdev;
3506 if (c4iw_fatal_error(rdev)) {
3508 close_complete_upcall(ep, -EIO);
3509 ep->com.state = DEAD;
3511 switch (ep->com.state) {
3520 ep->com.state = ABORTING;
3522 ep->com.state = CLOSING;
3525 * if we close before we see the fw4_ack() then we fix
3526 * up the timer state since we're reusing it.
3529 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3530 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3535 set_bit(CLOSE_SENT, &ep->com.flags);
3538 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3541 (void)stop_ep_timer(ep);
3542 ep->com.state = ABORTING;
3544 ep->com.state = MORIBUND;
3550 PDBG("%s ignoring disconnect ep %p state %u\n",
3551 __func__, ep, ep->com.state);
3560 set_bit(EP_DISC_ABORT, &ep->com.history);
3561 close_complete_upcall(ep, -ECONNRESET);
3562 ret = send_abort(ep);
3564 set_bit(EP_DISC_CLOSE, &ep->com.history);
3565 ret = send_halfclose(ep);
3568 set_bit(EP_DISC_FAIL, &ep->com.history);
3571 close_complete_upcall(ep, -EIO);
3574 struct c4iw_qp_attributes attrs;
3576 attrs.next_state = C4IW_QP_STATE_ERROR;
3577 ret = c4iw_modify_qp(ep->com.qp->rhp,
3579 C4IW_QP_ATTR_NEXT_STATE,
3583 "%s - qp <- error failed!\n",
3589 mutex_unlock(&ep->com.mutex);
3590 c4iw_put_ep(&ep->com);
3592 release_ep_resources(ep);
3596 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3597 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3600 int atid = be32_to_cpu(req->tid);
3602 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3603 (__force u32) req->tid);
3607 switch (req->retval) {
3609 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3610 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3611 send_fw_act_open_req(ep, atid);
3615 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3616 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3617 send_fw_act_open_req(ep, atid);
3622 pr_info("%s unexpected ofld conn wr retval %d\n",
3623 __func__, req->retval);
3626 pr_err("active ofld_connect_wr failure %d atid %d\n",
3628 mutex_lock(&dev->rdev.stats.lock);
3629 dev->rdev.stats.act_ofld_conn_fails++;
3630 mutex_unlock(&dev->rdev.stats.lock);
3631 connect_reply_upcall(ep, status2errno(req->retval));
3632 state_set(&ep->com, DEAD);
3633 if (ep->com.remote_addr.ss_family == AF_INET6) {
3634 struct sockaddr_in6 *sin6 =
3635 (struct sockaddr_in6 *)&ep->com.local_addr;
3636 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3637 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3639 remove_handle(dev, &dev->atid_idr, atid);
3640 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3641 dst_release(ep->dst);
3642 cxgb4_l2t_release(ep->l2t);
3643 c4iw_put_ep(&ep->com);
3646 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3647 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3649 struct sk_buff *rpl_skb;
3650 struct cpl_pass_accept_req *cpl;
3653 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3656 PDBG("%s passive open failure %d\n", __func__, req->retval);
3657 mutex_lock(&dev->rdev.stats.lock);
3658 dev->rdev.stats.pas_ofld_conn_fails++;
3659 mutex_unlock(&dev->rdev.stats.lock);
3662 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3663 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3664 (__force u32) htonl(
3665 (__force u32) req->tid)));
3666 ret = pass_accept_req(dev, rpl_skb);
3673 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3675 struct cpl_fw6_msg *rpl = cplhdr(skb);
3676 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3678 switch (rpl->type) {
3680 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3682 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3683 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3684 switch (req->t_state) {
3686 active_ofld_conn_reply(dev, skb, req);
3689 passive_ofld_conn_reply(dev, skb, req);
3692 pr_err("%s unexpected ofld conn wr state %d\n",
3693 __func__, req->t_state);
3701 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3704 __be16 hdr_len, vlantag, len;
3706 int tcp_hdr_len, ip_hdr_len;
3708 struct cpl_rx_pkt *cpl = cplhdr(skb);
3709 struct cpl_pass_accept_req *req;
3710 struct tcp_options_received tmp_opt;
3711 struct c4iw_dev *dev;
3712 enum chip_type type;
3714 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3715 /* Store values from cpl_rx_pkt in temporary location. */
3716 vlantag = cpl->vlan;
3718 l2info = cpl->l2info;
3719 hdr_len = cpl->hdr_len;
3722 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3725 * We need to parse the TCP options from SYN packet.
3726 * to generate cpl_pass_accept_req.
3728 memset(&tmp_opt, 0, sizeof(tmp_opt));
3729 tcp_clear_options(&tmp_opt);
3730 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3732 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3733 memset(req, 0, sizeof(*req));
3734 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3735 SYN_MAC_IDX_V(RX_MACIDX_G(
3736 be32_to_cpu(l2info))) |
3738 type = dev->rdev.lldi.adapter_type;
3739 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3740 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3742 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3743 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3744 eth_hdr_len = is_t4(type) ?
3745 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3746 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3747 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3748 IP_HDR_LEN_V(ip_hdr_len) |
3749 ETH_HDR_LEN_V(eth_hdr_len));
3750 } else { /* T6 and later */
3751 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3752 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3753 T6_IP_HDR_LEN_V(ip_hdr_len) |
3754 T6_ETH_HDR_LEN_V(eth_hdr_len));
3756 req->vlan = vlantag;
3758 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3759 PASS_OPEN_TOS_V(tos));
3760 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3761 if (tmp_opt.wscale_ok)
3762 req->tcpopt.wsf = tmp_opt.snd_wscale;
3763 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3764 if (tmp_opt.sack_ok)
3765 req->tcpopt.sack = 1;
3766 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3770 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3771 __be32 laddr, __be16 lport,
3772 __be32 raddr, __be16 rport,
3773 u32 rcv_isn, u32 filter, u16 window,
3774 u32 rss_qid, u8 port_id)
3776 struct sk_buff *req_skb;
3777 struct fw_ofld_connection_wr *req;
3778 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3781 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3782 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3783 memset(req, 0, sizeof(*req));
3784 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3785 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3786 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3787 req->le.filter = (__force __be32) filter;
3788 req->le.lport = lport;
3789 req->le.pport = rport;
3790 req->le.u.ipv4.lip = laddr;
3791 req->le.u.ipv4.pip = raddr;
3792 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3793 req->tcb.rcv_adv = htons(window);
3794 req->tcb.t_state_to_astid =
3795 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3796 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3797 FW_OFLD_CONNECTION_WR_ASTID_V(
3798 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3801 * We store the qid in opt2 which will be used by the firmware
3802 * to send us the wr response.
3804 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3807 * We initialize the MSS index in TCB to 0xF.
3808 * So that when driver sends cpl_pass_accept_rpl
3809 * TCB picks up the correct value. If this was 0
3810 * TP will ignore any value > 0 for MSS index.
3812 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3813 req->cookie = (uintptr_t)skb;
3815 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3816 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3818 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3826 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3827 * messages when a filter is being used instead of server to
3828 * redirect a syn packet. When packets hit filter they are redirected
3829 * to the offload queue and driver tries to establish the connection
3830 * using firmware work request.
3832 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3835 unsigned int filter;
3836 struct ethhdr *eh = NULL;
3837 struct vlan_ethhdr *vlan_eh = NULL;
3839 struct tcphdr *tcph;
3840 struct rss_header *rss = (void *)skb->data;
3841 struct cpl_rx_pkt *cpl = (void *)skb->data;
3842 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3843 struct l2t_entry *e;
3844 struct dst_entry *dst;
3845 struct c4iw_ep *lep = NULL;
3847 struct port_info *pi;
3848 struct net_device *pdev;
3849 u16 rss_qid, eth_hdr_len;
3852 struct neighbour *neigh;
3854 /* Drop all non-SYN packets */
3855 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3859 * Drop all packets which did not hit the filter.
3860 * Unlikely to happen.
3862 if (!(rss->filter_hit && rss->filter_tid))
3866 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3868 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3870 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3872 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3876 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3878 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3881 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3884 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3887 pr_err("T%d Chip is not supported\n",
3888 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3892 if (eth_hdr_len == ETH_HLEN) {
3893 eh = (struct ethhdr *)(req + 1);
3894 iph = (struct iphdr *)(eh + 1);
3896 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3897 iph = (struct iphdr *)(vlan_eh + 1);
3898 skb->vlan_tci = ntohs(cpl->vlan);
3901 if (iph->version != 0x4)
3904 tcph = (struct tcphdr *)(iph + 1);
3905 skb_set_network_header(skb, (void *)iph - (void *)rss);
3906 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3909 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3910 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3911 ntohs(tcph->source), iph->tos);
3913 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3914 iph->daddr, iph->saddr, tcph->dest,
3915 tcph->source, iph->tos);
3917 pr_err("%s - failed to find dst entry!\n",
3921 neigh = dst_neigh_lookup_skb(dst, skb);
3924 pr_err("%s - failed to allocate neigh!\n",
3929 if (neigh->dev->flags & IFF_LOOPBACK) {
3930 pdev = ip_dev_find(&init_net, iph->daddr);
3931 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3933 pi = (struct port_info *)netdev_priv(pdev);
3934 tx_chan = cxgb4_port_chan(pdev);
3937 pdev = get_real_dev(neigh->dev);
3938 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3940 pi = (struct port_info *)netdev_priv(pdev);
3941 tx_chan = cxgb4_port_chan(pdev);
3943 neigh_release(neigh);
3945 pr_err("%s - failed to allocate l2t entry!\n",
3950 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3951 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3952 window = (__force u16) htons((__force u16)tcph->window);
3954 /* Calcuate filter portion for LE region. */
3955 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3956 dev->rdev.lldi.ports[0],
3960 * Synthesize the cpl_pass_accept_req. We have everything except the
3961 * TID. Once firmware sends a reply with TID we update the TID field
3962 * in cpl and pass it through the regular cpl_pass_accept_req path.
3964 build_cpl_pass_accept_req(skb, stid, iph->tos);
3965 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3966 tcph->source, ntohl(tcph->seq), filter, window,
3967 rss_qid, pi->port_id);
3968 cxgb4_l2t_release(e);
3973 c4iw_put_ep(&lep->com);
3978 * These are the real handlers that are called from a
3981 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
3982 [CPL_ACT_ESTABLISH] = act_establish,
3983 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3984 [CPL_RX_DATA] = rx_data,
3985 [CPL_ABORT_RPL_RSS] = abort_rpl,
3986 [CPL_ABORT_RPL] = abort_rpl,
3987 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3988 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3989 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3990 [CPL_PASS_ESTABLISH] = pass_establish,
3991 [CPL_PEER_CLOSE] = peer_close,
3992 [CPL_ABORT_REQ_RSS] = peer_abort,
3993 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3994 [CPL_RDMA_TERMINATE] = terminate,
3995 [CPL_FW4_ACK] = fw4_ack,
3996 [CPL_FW6_MSG] = deferred_fw6_msg,
3997 [CPL_RX_PKT] = rx_pkt,
3998 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
3999 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4002 static void process_timeout(struct c4iw_ep *ep)
4004 struct c4iw_qp_attributes attrs;
4007 mutex_lock(&ep->com.mutex);
4008 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4010 set_bit(TIMEDOUT, &ep->com.history);
4011 switch (ep->com.state) {
4013 connect_reply_upcall(ep, -ETIMEDOUT);
4022 if (ep->com.cm_id && ep->com.qp) {
4023 attrs.next_state = C4IW_QP_STATE_ERROR;
4024 c4iw_modify_qp(ep->com.qp->rhp,
4025 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4028 close_complete_upcall(ep, -ETIMEDOUT);
4034 * These states are expected if the ep timed out at the same
4035 * time as another thread was calling stop_ep_timer().
4036 * So we silently do nothing for these states.
4041 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4042 __func__, ep, ep->hwtid, ep->com.state);
4045 mutex_unlock(&ep->com.mutex);
4047 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4048 c4iw_put_ep(&ep->com);
4051 static void process_timedout_eps(void)
4055 spin_lock_irq(&timeout_lock);
4056 while (!list_empty(&timeout_list)) {
4057 struct list_head *tmp;
4059 tmp = timeout_list.next;
4063 spin_unlock_irq(&timeout_lock);
4064 ep = list_entry(tmp, struct c4iw_ep, entry);
4065 process_timeout(ep);
4066 spin_lock_irq(&timeout_lock);
4068 spin_unlock_irq(&timeout_lock);
4071 static void process_work(struct work_struct *work)
4073 struct sk_buff *skb = NULL;
4074 struct c4iw_dev *dev;
4075 struct cpl_act_establish *rpl;
4076 unsigned int opcode;
4079 process_timedout_eps();
4080 while ((skb = skb_dequeue(&rxq))) {
4082 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4083 opcode = rpl->ot.opcode;
4085 BUG_ON(!work_handlers[opcode]);
4086 ret = work_handlers[opcode](dev, skb);
4089 process_timedout_eps();
4093 static DECLARE_WORK(skb_work, process_work);
4095 static void ep_timeout(unsigned long arg)
4097 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4100 spin_lock(&timeout_lock);
4101 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4103 * Only insert if it is not already on the list.
4105 if (!ep->entry.next) {
4106 list_add_tail(&ep->entry, &timeout_list);
4110 spin_unlock(&timeout_lock);
4112 queue_work(workq, &skb_work);
4116 * All the CM events are handled on a work queue to have a safe context.
4118 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4122 * Save dev in the skb->cb area.
4124 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4127 * Queue the skb and schedule the worker thread.
4129 skb_queue_tail(&rxq, skb);
4130 queue_work(workq, &skb_work);
4134 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4136 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4138 if (rpl->status != CPL_ERR_NONE) {
4139 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4140 "for tid %u\n", rpl->status, GET_TID(rpl));
4146 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4148 struct cpl_fw6_msg *rpl = cplhdr(skb);
4149 struct c4iw_wr_wait *wr_waitp;
4152 PDBG("%s type %u\n", __func__, rpl->type);
4154 switch (rpl->type) {
4155 case FW6_TYPE_WR_RPL:
4156 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4157 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4158 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4160 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4164 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4168 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4176 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4178 struct cpl_abort_req_rss *req = cplhdr(skb);
4180 unsigned int tid = GET_TID(req);
4182 ep = get_ep_from_tid(dev, tid);
4183 /* This EP will be dereferenced in peer_abort() */
4185 printk(KERN_WARNING MOD
4186 "Abort on non-existent endpoint, tid %d\n", tid);
4190 if (cxgb_is_neg_adv(req->status)) {
4191 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4192 __func__, ep->hwtid, req->status,
4193 neg_adv_str(req->status));
4196 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4199 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4206 * Most upcalls from the T4 Core go to sched() to
4207 * schedule the processing on a work queue.
4209 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4210 [CPL_ACT_ESTABLISH] = sched,
4211 [CPL_ACT_OPEN_RPL] = sched,
4212 [CPL_RX_DATA] = sched,
4213 [CPL_ABORT_RPL_RSS] = sched,
4214 [CPL_ABORT_RPL] = sched,
4215 [CPL_PASS_OPEN_RPL] = sched,
4216 [CPL_CLOSE_LISTSRV_RPL] = sched,
4217 [CPL_PASS_ACCEPT_REQ] = sched,
4218 [CPL_PASS_ESTABLISH] = sched,
4219 [CPL_PEER_CLOSE] = sched,
4220 [CPL_CLOSE_CON_RPL] = sched,
4221 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4222 [CPL_RDMA_TERMINATE] = sched,
4223 [CPL_FW4_ACK] = sched,
4224 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4225 [CPL_FW6_MSG] = fw6_msg,
4226 [CPL_RX_PKT] = sched
4229 int __init c4iw_cm_init(void)
4231 spin_lock_init(&timeout_lock);
4232 skb_queue_head_init(&rxq);
4234 workq = create_singlethread_workqueue("iw_cxgb4");
4241 void c4iw_cm_term(void)
4243 WARN_ON(!list_empty(&timeout_list));
4244 flush_workqueue(workq);
4245 destroy_workqueue(workq);