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 cpl_rx_data_ack *req;
1358 struct sk_buff *skb;
1359 int wrlen = roundup(sizeof *req, 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 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1377 memset(req, 0, wrlen);
1378 INIT_TP_WR(req, ep->hwtid);
1379 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1381 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
1383 RX_DACK_MODE_V(dack_mode));
1384 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1385 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1389 #define RELAXED_IRD_NEGOTIATION 1
1392 * process_mpa_reply - process streaming mode MPA reply
1396 * 0 upon success indicating a connect request was delivered to the ULP
1397 * or the mpa request is incomplete but valid so far.
1399 * 1 if a failure requires the caller to close the connection.
1401 * 2 if a failure requires the caller to abort the connection.
1403 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1405 struct mpa_message *mpa;
1406 struct mpa_v2_conn_params *mpa_v2_params;
1408 u16 resp_ird, resp_ord;
1409 u8 rtr_mismatch = 0, insuff_ird = 0;
1410 struct c4iw_qp_attributes attrs;
1411 enum c4iw_qp_attr_mask mask;
1415 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1418 * If we get more than the supported amount of private data
1419 * then we must fail this connection.
1421 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1423 goto err_stop_timer;
1427 * copy the new data into our accumulation buffer.
1429 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1431 ep->mpa_pkt_len += skb->len;
1434 * if we don't even have the mpa message, then bail.
1436 if (ep->mpa_pkt_len < sizeof(*mpa))
1438 mpa = (struct mpa_message *) ep->mpa_pkt;
1440 /* Validate MPA header. */
1441 if (mpa->revision > mpa_rev) {
1442 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1443 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1445 goto err_stop_timer;
1447 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1449 goto err_stop_timer;
1452 plen = ntohs(mpa->private_data_size);
1455 * Fail if there's too much private data.
1457 if (plen > MPA_MAX_PRIVATE_DATA) {
1459 goto err_stop_timer;
1463 * If plen does not account for pkt size
1465 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1467 goto err_stop_timer;
1470 ep->plen = (u8) plen;
1473 * If we don't have all the pdata yet, then bail.
1474 * We'll continue process when more data arrives.
1476 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1479 if (mpa->flags & MPA_REJECT) {
1480 err = -ECONNREFUSED;
1481 goto err_stop_timer;
1485 * Stop mpa timer. If it expired, then
1486 * we ignore the MPA reply. process_timeout()
1487 * will abort the connection.
1489 if (stop_ep_timer(ep))
1493 * If we get here we have accumulated the entire mpa
1494 * start reply message including private data. And
1495 * the MPA header is valid.
1497 __state_set(&ep->com, FPDU_MODE);
1498 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1499 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1500 ep->mpa_attr.version = mpa->revision;
1501 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1503 if (mpa->revision == 2) {
1504 ep->mpa_attr.enhanced_rdma_conn =
1505 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1506 if (ep->mpa_attr.enhanced_rdma_conn) {
1507 mpa_v2_params = (struct mpa_v2_conn_params *)
1508 (ep->mpa_pkt + sizeof(*mpa));
1509 resp_ird = ntohs(mpa_v2_params->ird) &
1510 MPA_V2_IRD_ORD_MASK;
1511 resp_ord = ntohs(mpa_v2_params->ord) &
1512 MPA_V2_IRD_ORD_MASK;
1513 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1514 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1517 * This is a double-check. Ideally, below checks are
1518 * not required since ird/ord stuff has been taken
1519 * care of in c4iw_accept_cr
1521 if (ep->ird < resp_ord) {
1522 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1523 ep->com.dev->rdev.lldi.max_ordird_qp)
1527 } else if (ep->ird > resp_ord) {
1530 if (ep->ord > resp_ird) {
1531 if (RELAXED_IRD_NEGOTIATION)
1542 if (ntohs(mpa_v2_params->ird) &
1543 MPA_V2_PEER2PEER_MODEL) {
1544 if (ntohs(mpa_v2_params->ord) &
1545 MPA_V2_RDMA_WRITE_RTR)
1546 ep->mpa_attr.p2p_type =
1547 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1548 else if (ntohs(mpa_v2_params->ord) &
1549 MPA_V2_RDMA_READ_RTR)
1550 ep->mpa_attr.p2p_type =
1551 FW_RI_INIT_P2PTYPE_READ_REQ;
1554 } else if (mpa->revision == 1)
1556 ep->mpa_attr.p2p_type = p2p_type;
1558 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1559 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1560 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1561 ep->mpa_attr.recv_marker_enabled,
1562 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1563 ep->mpa_attr.p2p_type, p2p_type);
1566 * If responder's RTR does not match with that of initiator, assign
1567 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1568 * generated when moving QP to RTS state.
1569 * A TERM message will be sent after QP has moved to RTS state
1571 if ((ep->mpa_attr.version == 2) && peer2peer &&
1572 (ep->mpa_attr.p2p_type != p2p_type)) {
1573 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1577 attrs.mpa_attr = ep->mpa_attr;
1578 attrs.max_ird = ep->ird;
1579 attrs.max_ord = ep->ord;
1580 attrs.llp_stream_handle = ep;
1581 attrs.next_state = C4IW_QP_STATE_RTS;
1583 mask = C4IW_QP_ATTR_NEXT_STATE |
1584 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1585 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1587 /* bind QP and TID with INIT_WR */
1588 err = c4iw_modify_qp(ep->com.qp->rhp,
1589 ep->com.qp, mask, &attrs, 1);
1594 * If responder's RTR requirement did not match with what initiator
1595 * supports, generate TERM message
1598 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1599 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1600 attrs.ecode = MPA_NOMATCH_RTR;
1601 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1602 attrs.send_term = 1;
1603 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1604 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1611 * Generate TERM if initiator IRD is not sufficient for responder
1612 * provided ORD. Currently, we do the same behaviour even when
1613 * responder provided IRD is also not sufficient as regards to
1617 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1619 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1620 attrs.ecode = MPA_INSUFF_IRD;
1621 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1622 attrs.send_term = 1;
1623 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1624 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1635 connect_reply_upcall(ep, err);
1640 * process_mpa_request - process streaming mode MPA request
1644 * 0 upon success indicating a connect request was delivered to the ULP
1645 * or the mpa request is incomplete but valid so far.
1647 * 1 if a failure requires the caller to close the connection.
1649 * 2 if a failure requires the caller to abort the connection.
1651 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1653 struct mpa_message *mpa;
1654 struct mpa_v2_conn_params *mpa_v2_params;
1657 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1660 * If we get more than the supported amount of private data
1661 * then we must fail this connection.
1663 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1664 goto err_stop_timer;
1666 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1669 * Copy the new data into our accumulation buffer.
1671 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1673 ep->mpa_pkt_len += skb->len;
1676 * If we don't even have the mpa message, then bail.
1677 * We'll continue process when more data arrives.
1679 if (ep->mpa_pkt_len < sizeof(*mpa))
1682 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1683 mpa = (struct mpa_message *) ep->mpa_pkt;
1686 * Validate MPA Header.
1688 if (mpa->revision > mpa_rev) {
1689 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1690 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1691 goto err_stop_timer;
1694 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1695 goto err_stop_timer;
1697 plen = ntohs(mpa->private_data_size);
1700 * Fail if there's too much private data.
1702 if (plen > MPA_MAX_PRIVATE_DATA)
1703 goto err_stop_timer;
1706 * If plen does not account for pkt size
1708 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1709 goto err_stop_timer;
1710 ep->plen = (u8) plen;
1713 * If we don't have all the pdata yet, then bail.
1715 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1719 * If we get here we have accumulated the entire mpa
1720 * start reply message including private data.
1722 ep->mpa_attr.initiator = 0;
1723 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1724 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1725 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1726 ep->mpa_attr.version = mpa->revision;
1727 if (mpa->revision == 1)
1728 ep->tried_with_mpa_v1 = 1;
1729 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1731 if (mpa->revision == 2) {
1732 ep->mpa_attr.enhanced_rdma_conn =
1733 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1734 if (ep->mpa_attr.enhanced_rdma_conn) {
1735 mpa_v2_params = (struct mpa_v2_conn_params *)
1736 (ep->mpa_pkt + sizeof(*mpa));
1737 ep->ird = ntohs(mpa_v2_params->ird) &
1738 MPA_V2_IRD_ORD_MASK;
1739 ep->ird = min_t(u32, ep->ird,
1740 cur_max_read_depth(ep->com.dev));
1741 ep->ord = ntohs(mpa_v2_params->ord) &
1742 MPA_V2_IRD_ORD_MASK;
1743 ep->ord = min_t(u32, ep->ord,
1744 cur_max_read_depth(ep->com.dev));
1745 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1747 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1749 if (ntohs(mpa_v2_params->ord) &
1750 MPA_V2_RDMA_WRITE_RTR)
1751 ep->mpa_attr.p2p_type =
1752 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1753 else if (ntohs(mpa_v2_params->ord) &
1754 MPA_V2_RDMA_READ_RTR)
1755 ep->mpa_attr.p2p_type =
1756 FW_RI_INIT_P2PTYPE_READ_REQ;
1759 } else if (mpa->revision == 1)
1761 ep->mpa_attr.p2p_type = p2p_type;
1763 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1764 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1765 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1766 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1767 ep->mpa_attr.p2p_type);
1769 __state_set(&ep->com, MPA_REQ_RCVD);
1772 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1773 if (ep->parent_ep->com.state != DEAD) {
1774 if (connect_request_upcall(ep))
1775 goto err_unlock_parent;
1777 goto err_unlock_parent;
1779 mutex_unlock(&ep->parent_ep->com.mutex);
1783 mutex_unlock(&ep->parent_ep->com.mutex);
1786 (void)stop_ep_timer(ep);
1791 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1794 struct cpl_rx_data *hdr = cplhdr(skb);
1795 unsigned int dlen = ntohs(hdr->len);
1796 unsigned int tid = GET_TID(hdr);
1797 __u8 status = hdr->status;
1800 ep = get_ep_from_tid(dev, tid);
1803 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1804 skb_pull(skb, sizeof(*hdr));
1805 skb_trim(skb, dlen);
1806 mutex_lock(&ep->com.mutex);
1808 /* update RX credits */
1809 update_rx_credits(ep, dlen);
1811 switch (ep->com.state) {
1813 ep->rcv_seq += dlen;
1814 disconnect = process_mpa_reply(ep, skb);
1817 ep->rcv_seq += dlen;
1818 disconnect = process_mpa_request(ep, skb);
1821 struct c4iw_qp_attributes attrs;
1822 BUG_ON(!ep->com.qp);
1824 pr_err("%s Unexpected streaming data." \
1825 " qpid %u ep %p state %d tid %u status %d\n",
1826 __func__, ep->com.qp->wq.sq.qid, ep,
1827 ep->com.state, ep->hwtid, status);
1828 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1829 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1830 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1837 mutex_unlock(&ep->com.mutex);
1839 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1840 c4iw_put_ep(&ep->com);
1844 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1847 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1849 unsigned int tid = GET_TID(rpl);
1851 ep = get_ep_from_tid(dev, tid);
1853 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1856 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1857 mutex_lock(&ep->com.mutex);
1858 switch (ep->com.state) {
1860 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1861 __state_set(&ep->com, DEAD);
1865 printk(KERN_ERR "%s ep %p state %d\n",
1866 __func__, ep, ep->com.state);
1869 mutex_unlock(&ep->com.mutex);
1872 release_ep_resources(ep);
1873 c4iw_put_ep(&ep->com);
1877 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1879 struct sk_buff *skb;
1880 struct fw_ofld_connection_wr *req;
1881 unsigned int mtu_idx;
1883 struct sockaddr_in *sin;
1886 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1887 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1888 memset(req, 0, sizeof(*req));
1889 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1890 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1891 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1892 ep->com.dev->rdev.lldi.ports[0],
1894 sin = (struct sockaddr_in *)&ep->com.local_addr;
1895 req->le.lport = sin->sin_port;
1896 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1897 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1898 req->le.pport = sin->sin_port;
1899 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1900 req->tcb.t_state_to_astid =
1901 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1902 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1903 req->tcb.cplrxdataack_cplpassacceptrpl =
1904 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1905 req->tcb.tx_max = (__force __be32) jiffies;
1906 req->tcb.rcv_adv = htons(1);
1907 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1908 enable_tcp_timestamps,
1909 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1910 wscale = cxgb_compute_wscale(rcv_win);
1913 * Specify the largest window that will fit in opt0. The
1914 * remainder will be specified in the rx_data_ack.
1916 win = ep->rcv_win >> 10;
1917 if (win > RCV_BUFSIZ_M)
1920 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1921 (nocong ? NO_CONG_F : 0) |
1924 WND_SCALE_V(wscale) |
1925 MSS_IDX_V(mtu_idx) |
1926 L2T_IDX_V(ep->l2t->idx) |
1927 TX_CHAN_V(ep->tx_chan) |
1928 SMAC_SEL_V(ep->smac_idx) |
1929 DSCP_V(ep->tos >> 2) |
1930 ULP_MODE_V(ULP_MODE_TCPDDP) |
1932 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1933 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1935 CCTRL_ECN_V(enable_ecn) |
1936 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1937 if (enable_tcp_timestamps)
1938 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1939 if (enable_tcp_sack)
1940 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1941 if (wscale && enable_tcp_window_scaling)
1942 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1943 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1944 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1945 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1946 set_bit(ACT_OFLD_CONN, &ep->com.history);
1947 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1951 * Some of the error codes above implicitly indicate that there is no TID
1952 * allocated with the result of an ACT_OPEN. We use this predicate to make
1955 static inline int act_open_has_tid(int status)
1957 return (status != CPL_ERR_TCAM_PARITY &&
1958 status != CPL_ERR_TCAM_MISS &&
1959 status != CPL_ERR_TCAM_FULL &&
1960 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1961 status != CPL_ERR_CONN_EXIST);
1964 static char *neg_adv_str(unsigned int status)
1967 case CPL_ERR_RTX_NEG_ADVICE:
1968 return "Retransmit timeout";
1969 case CPL_ERR_PERSIST_NEG_ADVICE:
1970 return "Persist timeout";
1971 case CPL_ERR_KEEPALV_NEG_ADVICE:
1972 return "Keepalive timeout";
1978 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1980 ep->snd_win = snd_win;
1981 ep->rcv_win = rcv_win;
1982 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
1985 #define ACT_OPEN_RETRY_COUNT 2
1987 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1988 struct dst_entry *dst, struct c4iw_dev *cdev,
1989 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
1991 struct neighbour *n;
1993 struct net_device *pdev;
1995 n = dst_neigh_lookup(dst, peer_ip);
2001 if (n->dev->flags & IFF_LOOPBACK) {
2003 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2004 else if (IS_ENABLED(CONFIG_IPV6))
2005 for_each_netdev(&init_net, pdev) {
2006 if (ipv6_chk_addr(&init_net,
2007 (struct in6_addr *)peer_ip,
2018 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2019 n, pdev, rt_tos2priority(tos));
2022 ep->mtu = pdev->mtu;
2023 ep->tx_chan = cxgb4_port_chan(pdev);
2024 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2025 cxgb4_port_viid(pdev));
2026 step = cdev->rdev.lldi.ntxq /
2027 cdev->rdev.lldi.nchan;
2028 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2029 step = cdev->rdev.lldi.nrxq /
2030 cdev->rdev.lldi.nchan;
2031 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2032 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2033 cxgb4_port_idx(pdev) * step];
2034 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2037 pdev = get_real_dev(n->dev);
2038 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2042 ep->mtu = dst_mtu(dst);
2043 ep->tx_chan = cxgb4_port_chan(pdev);
2044 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2045 cxgb4_port_viid(pdev));
2046 step = cdev->rdev.lldi.ntxq /
2047 cdev->rdev.lldi.nchan;
2048 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2049 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2050 step = cdev->rdev.lldi.nrxq /
2051 cdev->rdev.lldi.nchan;
2052 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2053 cxgb4_port_idx(pdev) * step];
2054 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2057 ep->retry_with_mpa_v1 = 0;
2058 ep->tried_with_mpa_v1 = 0;
2070 static int c4iw_reconnect(struct c4iw_ep *ep)
2074 struct sockaddr_in *laddr = (struct sockaddr_in *)
2075 &ep->com.cm_id->m_local_addr;
2076 struct sockaddr_in *raddr = (struct sockaddr_in *)
2077 &ep->com.cm_id->m_remote_addr;
2078 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2079 &ep->com.cm_id->m_local_addr;
2080 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2081 &ep->com.cm_id->m_remote_addr;
2085 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2086 init_timer(&ep->timer);
2087 c4iw_init_wr_wait(&ep->com.wr_wait);
2089 /* When MPA revision is different on nodes, the node with MPA_rev=2
2090 * tries to reconnect with MPA_rev 1 for the same EP through
2091 * c4iw_reconnect(), where the same EP is assigned with new tid for
2092 * further connection establishment. As we are using the same EP pointer
2093 * for reconnect, few skbs are used during the previous c4iw_connect(),
2094 * which leaves the EP with inadequate skbs for further
2095 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2096 * skb_list() during peer_abort(). Allocate skbs which is already used.
2098 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2099 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2105 * Allocate an active TID to initiate a TCP connection.
2107 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2108 if (ep->atid == -1) {
2109 pr_err("%s - cannot alloc atid.\n", __func__);
2113 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2116 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2117 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2118 laddr->sin_addr.s_addr,
2119 raddr->sin_addr.s_addr,
2121 raddr->sin_port, ep->com.cm_id->tos);
2123 ra = (__u8 *)&raddr->sin_addr;
2125 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2127 laddr6->sin6_addr.s6_addr,
2128 raddr6->sin6_addr.s6_addr,
2130 raddr6->sin6_port, 0,
2131 raddr6->sin6_scope_id);
2133 ra = (__u8 *)&raddr6->sin6_addr;
2136 pr_err("%s - cannot find route.\n", __func__);
2137 err = -EHOSTUNREACH;
2140 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2141 ep->com.dev->rdev.lldi.adapter_type,
2142 ep->com.cm_id->tos);
2144 pr_err("%s - cannot alloc l2e.\n", __func__);
2148 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2149 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2152 state_set(&ep->com, CONNECTING);
2153 ep->tos = ep->com.cm_id->tos;
2155 /* send connect request to rnic */
2156 err = send_connect(ep);
2160 cxgb4_l2t_release(ep->l2t);
2162 dst_release(ep->dst);
2164 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2165 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2168 * remember to send notification to upper layer.
2169 * We are in here so the upper layer is not aware that this is
2170 * re-connect attempt and so, upper layer is still waiting for
2171 * response of 1st connect request.
2173 connect_reply_upcall(ep, -ECONNRESET);
2175 c4iw_put_ep(&ep->com);
2180 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2183 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2184 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2185 ntohl(rpl->atid_status)));
2186 struct tid_info *t = dev->rdev.lldi.tids;
2187 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2188 struct sockaddr_in *la;
2189 struct sockaddr_in *ra;
2190 struct sockaddr_in6 *la6;
2191 struct sockaddr_in6 *ra6;
2194 ep = lookup_atid(t, atid);
2195 la = (struct sockaddr_in *)&ep->com.local_addr;
2196 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2197 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2198 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2200 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2201 status, status2errno(status));
2203 if (cxgb_is_neg_adv(status)) {
2204 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2205 __func__, atid, status, neg_adv_str(status));
2206 ep->stats.connect_neg_adv++;
2207 mutex_lock(&dev->rdev.stats.lock);
2208 dev->rdev.stats.neg_adv++;
2209 mutex_unlock(&dev->rdev.stats.lock);
2213 set_bit(ACT_OPEN_RPL, &ep->com.history);
2216 * Log interesting failures.
2219 case CPL_ERR_CONN_RESET:
2220 case CPL_ERR_CONN_TIMEDOUT:
2222 case CPL_ERR_TCAM_FULL:
2223 mutex_lock(&dev->rdev.stats.lock);
2224 dev->rdev.stats.tcam_full++;
2225 mutex_unlock(&dev->rdev.stats.lock);
2226 if (ep->com.local_addr.ss_family == AF_INET &&
2227 dev->rdev.lldi.enable_fw_ofld_conn) {
2228 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2229 ntohl(rpl->atid_status))));
2235 case CPL_ERR_CONN_EXIST:
2236 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2237 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2238 if (ep->com.remote_addr.ss_family == AF_INET6) {
2239 struct sockaddr_in6 *sin6 =
2240 (struct sockaddr_in6 *)
2241 &ep->com.local_addr;
2243 ep->com.dev->rdev.lldi.ports[0],
2245 &sin6->sin6_addr.s6_addr, 1);
2247 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2249 cxgb4_free_atid(t, atid);
2250 dst_release(ep->dst);
2251 cxgb4_l2t_release(ep->l2t);
2257 if (ep->com.local_addr.ss_family == AF_INET) {
2258 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2259 atid, status, status2errno(status),
2260 &la->sin_addr.s_addr, ntohs(la->sin_port),
2261 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2263 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2264 atid, status, status2errno(status),
2265 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2266 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2272 connect_reply_upcall(ep, status2errno(status));
2273 state_set(&ep->com, DEAD);
2275 if (ep->com.remote_addr.ss_family == AF_INET6) {
2276 struct sockaddr_in6 *sin6 =
2277 (struct sockaddr_in6 *)&ep->com.local_addr;
2278 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2279 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2281 if (status && act_open_has_tid(status))
2282 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2284 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2285 cxgb4_free_atid(t, atid);
2286 dst_release(ep->dst);
2287 cxgb4_l2t_release(ep->l2t);
2288 c4iw_put_ep(&ep->com);
2293 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2295 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2296 unsigned int stid = GET_TID(rpl);
2297 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2300 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2303 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2304 rpl->status, status2errno(rpl->status));
2305 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2306 c4iw_put_ep(&ep->com);
2311 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2313 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2314 unsigned int stid = GET_TID(rpl);
2315 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2317 PDBG("%s ep %p\n", __func__, ep);
2318 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2319 c4iw_put_ep(&ep->com);
2323 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2324 struct cpl_pass_accept_req *req)
2326 struct cpl_pass_accept_rpl *rpl;
2327 unsigned int mtu_idx;
2331 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2333 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2335 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2336 BUG_ON(skb_cloned(skb));
2340 if (!is_t4(adapter_type)) {
2341 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2343 INIT_TP_WR(rpl5, ep->hwtid);
2345 skb_trim(skb, sizeof(*rpl));
2346 INIT_TP_WR(rpl, ep->hwtid);
2348 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2351 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2352 enable_tcp_timestamps && req->tcpopt.tstamp,
2353 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2354 wscale = cxgb_compute_wscale(rcv_win);
2357 * Specify the largest window that will fit in opt0. The
2358 * remainder will be specified in the rx_data_ack.
2360 win = ep->rcv_win >> 10;
2361 if (win > RCV_BUFSIZ_M)
2363 opt0 = (nocong ? NO_CONG_F : 0) |
2366 WND_SCALE_V(wscale) |
2367 MSS_IDX_V(mtu_idx) |
2368 L2T_IDX_V(ep->l2t->idx) |
2369 TX_CHAN_V(ep->tx_chan) |
2370 SMAC_SEL_V(ep->smac_idx) |
2371 DSCP_V(ep->tos >> 2) |
2372 ULP_MODE_V(ULP_MODE_TCPDDP) |
2374 opt2 = RX_CHANNEL_V(0) |
2375 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2377 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2378 opt2 |= TSTAMPS_EN_F;
2379 if (enable_tcp_sack && req->tcpopt.sack)
2381 if (wscale && enable_tcp_window_scaling)
2382 opt2 |= WND_SCALE_EN_F;
2384 const struct tcphdr *tcph;
2385 u32 hlen = ntohl(req->hdr_len);
2387 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2388 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2391 tcph = (const void *)(req + 1) +
2392 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2393 if (tcph->ece && tcph->cwr)
2394 opt2 |= CCTRL_ECN_V(1);
2396 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2397 u32 isn = (prandom_u32() & ~7UL) - 1;
2398 opt2 |= T5_OPT_2_VALID_F;
2399 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2402 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2405 rpl5->iss = cpu_to_be32(isn);
2406 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2409 rpl->opt0 = cpu_to_be64(opt0);
2410 rpl->opt2 = cpu_to_be32(opt2);
2411 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2412 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2414 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2417 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2419 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2420 BUG_ON(skb_cloned(skb));
2421 skb_trim(skb, sizeof(struct cpl_tid_release));
2422 release_tid(&dev->rdev, hwtid, skb);
2426 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2428 struct c4iw_ep *child_ep = NULL, *parent_ep;
2429 struct cpl_pass_accept_req *req = cplhdr(skb);
2430 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2431 struct tid_info *t = dev->rdev.lldi.tids;
2432 unsigned int hwtid = GET_TID(req);
2433 struct dst_entry *dst;
2434 __u8 local_ip[16], peer_ip[16];
2435 __be16 local_port, peer_port;
2436 struct sockaddr_in6 *sin6;
2438 u16 peer_mss = ntohs(req->tcpopt.mss);
2440 unsigned short hdrs;
2441 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2443 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2445 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2449 if (state_read(&parent_ep->com) != LISTEN) {
2450 PDBG("%s - listening ep not in LISTEN\n", __func__);
2454 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2455 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2457 /* Find output route */
2459 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2460 , __func__, parent_ep, hwtid,
2461 local_ip, peer_ip, ntohs(local_port),
2462 ntohs(peer_port), peer_mss);
2463 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2464 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2465 local_port, peer_port, tos);
2467 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2468 , __func__, parent_ep, hwtid,
2469 local_ip, peer_ip, ntohs(local_port),
2470 ntohs(peer_port), peer_mss);
2471 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2472 local_ip, peer_ip, local_port, peer_port,
2473 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2474 ((struct sockaddr_in6 *)
2475 &parent_ep->com.local_addr)->sin6_scope_id);
2478 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2483 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2485 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2491 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2492 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2494 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2501 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2502 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2503 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2504 child_ep->mtu = peer_mss + hdrs;
2506 skb_queue_head_init(&child_ep->com.ep_skb_list);
2507 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2510 state_set(&child_ep->com, CONNECTING);
2511 child_ep->com.dev = dev;
2512 child_ep->com.cm_id = NULL;
2515 struct sockaddr_in *sin = (struct sockaddr_in *)
2516 &child_ep->com.local_addr;
2518 sin->sin_family = PF_INET;
2519 sin->sin_port = local_port;
2520 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2522 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2523 sin->sin_family = PF_INET;
2524 sin->sin_port = ((struct sockaddr_in *)
2525 &parent_ep->com.local_addr)->sin_port;
2526 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2528 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2529 sin->sin_family = PF_INET;
2530 sin->sin_port = peer_port;
2531 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2533 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2534 sin6->sin6_family = PF_INET6;
2535 sin6->sin6_port = local_port;
2536 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2538 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2539 sin6->sin6_family = PF_INET6;
2540 sin6->sin6_port = ((struct sockaddr_in6 *)
2541 &parent_ep->com.local_addr)->sin6_port;
2542 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2544 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2545 sin6->sin6_family = PF_INET6;
2546 sin6->sin6_port = peer_port;
2547 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2550 c4iw_get_ep(&parent_ep->com);
2551 child_ep->parent_ep = parent_ep;
2552 child_ep->tos = tos;
2553 child_ep->dst = dst;
2554 child_ep->hwtid = hwtid;
2556 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2557 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2559 init_timer(&child_ep->timer);
2560 cxgb4_insert_tid(t, child_ep, hwtid);
2561 insert_ep_tid(child_ep);
2562 if (accept_cr(child_ep, skb, req)) {
2563 c4iw_put_ep(&parent_ep->com);
2564 release_ep_resources(child_ep);
2566 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2569 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2570 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2571 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2575 c4iw_put_ep(&child_ep->com);
2577 reject_cr(dev, hwtid, skb);
2579 c4iw_put_ep(&parent_ep->com);
2584 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2587 struct cpl_pass_establish *req = cplhdr(skb);
2588 unsigned int tid = GET_TID(req);
2591 ep = get_ep_from_tid(dev, tid);
2592 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2593 ep->snd_seq = be32_to_cpu(req->snd_isn);
2594 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2596 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2597 ntohs(req->tcp_opt));
2599 set_emss(ep, ntohs(req->tcp_opt));
2601 dst_confirm(ep->dst);
2602 mutex_lock(&ep->com.mutex);
2603 ep->com.state = MPA_REQ_WAIT;
2605 set_bit(PASS_ESTAB, &ep->com.history);
2606 ret = send_flowc(ep);
2607 mutex_unlock(&ep->com.mutex);
2609 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2610 c4iw_put_ep(&ep->com);
2615 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2617 struct cpl_peer_close *hdr = cplhdr(skb);
2619 struct c4iw_qp_attributes attrs;
2622 unsigned int tid = GET_TID(hdr);
2625 ep = get_ep_from_tid(dev, tid);
2629 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2630 dst_confirm(ep->dst);
2632 set_bit(PEER_CLOSE, &ep->com.history);
2633 mutex_lock(&ep->com.mutex);
2634 switch (ep->com.state) {
2636 __state_set(&ep->com, CLOSING);
2639 __state_set(&ep->com, CLOSING);
2640 connect_reply_upcall(ep, -ECONNRESET);
2645 * We're gonna mark this puppy DEAD, but keep
2646 * the reference on it until the ULP accepts or
2647 * rejects the CR. Also wake up anyone waiting
2648 * in rdma connection migration (see c4iw_accept_cr()).
2650 __state_set(&ep->com, CLOSING);
2651 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2652 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2655 __state_set(&ep->com, CLOSING);
2656 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2657 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2661 __state_set(&ep->com, CLOSING);
2662 attrs.next_state = C4IW_QP_STATE_CLOSING;
2663 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2664 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2665 if (ret != -ECONNRESET) {
2666 peer_close_upcall(ep);
2674 __state_set(&ep->com, MORIBUND);
2678 (void)stop_ep_timer(ep);
2679 if (ep->com.cm_id && ep->com.qp) {
2680 attrs.next_state = C4IW_QP_STATE_IDLE;
2681 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2682 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2684 close_complete_upcall(ep, 0);
2685 __state_set(&ep->com, DEAD);
2695 mutex_unlock(&ep->com.mutex);
2697 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2699 release_ep_resources(ep);
2700 c4iw_put_ep(&ep->com);
2704 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2706 struct cpl_abort_req_rss *req = cplhdr(skb);
2708 struct cpl_abort_rpl *rpl;
2709 struct sk_buff *rpl_skb;
2710 struct c4iw_qp_attributes attrs;
2713 unsigned int tid = GET_TID(req);
2715 ep = get_ep_from_tid(dev, tid);
2719 if (cxgb_is_neg_adv(req->status)) {
2720 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2721 __func__, ep->hwtid, req->status,
2722 neg_adv_str(req->status));
2723 ep->stats.abort_neg_adv++;
2724 mutex_lock(&dev->rdev.stats.lock);
2725 dev->rdev.stats.neg_adv++;
2726 mutex_unlock(&dev->rdev.stats.lock);
2729 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2731 set_bit(PEER_ABORT, &ep->com.history);
2734 * Wake up any threads in rdma_init() or rdma_fini().
2735 * However, this is not needed if com state is just
2738 if (ep->com.state != MPA_REQ_SENT)
2739 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2741 mutex_lock(&ep->com.mutex);
2742 switch (ep->com.state) {
2744 c4iw_put_ep(&ep->parent_ep->com);
2747 (void)stop_ep_timer(ep);
2750 (void)stop_ep_timer(ep);
2751 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2752 connect_reply_upcall(ep, -ECONNRESET);
2755 * we just don't send notification upwards because we
2756 * want to retry with mpa_v1 without upper layers even
2759 * do some housekeeping so as to re-initiate the
2762 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2764 ep->retry_with_mpa_v1 = 1;
2776 if (ep->com.cm_id && ep->com.qp) {
2777 attrs.next_state = C4IW_QP_STATE_ERROR;
2778 ret = c4iw_modify_qp(ep->com.qp->rhp,
2779 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2783 "%s - qp <- error failed!\n",
2786 peer_abort_upcall(ep);
2791 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2792 mutex_unlock(&ep->com.mutex);
2798 dst_confirm(ep->dst);
2799 if (ep->com.state != ABORTING) {
2800 __state_set(&ep->com, DEAD);
2801 /* we don't release if we want to retry with mpa_v1 */
2802 if (!ep->retry_with_mpa_v1)
2805 mutex_unlock(&ep->com.mutex);
2807 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2808 if (WARN_ON(!rpl_skb)) {
2812 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2813 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2814 INIT_TP_WR(rpl, ep->hwtid);
2815 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2816 rpl->cmd = CPL_ABORT_NO_RST;
2817 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2820 release_ep_resources(ep);
2821 else if (ep->retry_with_mpa_v1) {
2822 if (ep->com.remote_addr.ss_family == AF_INET6) {
2823 struct sockaddr_in6 *sin6 =
2824 (struct sockaddr_in6 *)
2825 &ep->com.local_addr;
2827 ep->com.dev->rdev.lldi.ports[0],
2828 (const u32 *)&sin6->sin6_addr.s6_addr,
2831 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2832 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2833 dst_release(ep->dst);
2834 cxgb4_l2t_release(ep->l2t);
2839 c4iw_put_ep(&ep->com);
2840 /* Dereferencing ep, referenced in peer_abort_intr() */
2841 c4iw_put_ep(&ep->com);
2845 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2848 struct c4iw_qp_attributes attrs;
2849 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2851 unsigned int tid = GET_TID(rpl);
2853 ep = get_ep_from_tid(dev, tid);
2857 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2860 /* The cm_id may be null if we failed to connect */
2861 mutex_lock(&ep->com.mutex);
2862 set_bit(CLOSE_CON_RPL, &ep->com.history);
2863 switch (ep->com.state) {
2865 __state_set(&ep->com, MORIBUND);
2868 (void)stop_ep_timer(ep);
2869 if ((ep->com.cm_id) && (ep->com.qp)) {
2870 attrs.next_state = C4IW_QP_STATE_IDLE;
2871 c4iw_modify_qp(ep->com.qp->rhp,
2873 C4IW_QP_ATTR_NEXT_STATE,
2876 close_complete_upcall(ep, 0);
2877 __state_set(&ep->com, DEAD);
2887 mutex_unlock(&ep->com.mutex);
2889 release_ep_resources(ep);
2890 c4iw_put_ep(&ep->com);
2894 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2896 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2897 unsigned int tid = GET_TID(rpl);
2899 struct c4iw_qp_attributes attrs;
2901 ep = get_ep_from_tid(dev, tid);
2904 if (ep && ep->com.qp) {
2905 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2906 ep->com.qp->wq.sq.qid);
2907 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2908 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2909 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2911 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2912 c4iw_put_ep(&ep->com);
2918 * Upcall from the adapter indicating data has been transmitted.
2919 * For us its just the single MPA request or reply. We can now free
2920 * the skb holding the mpa message.
2922 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2925 struct cpl_fw4_ack *hdr = cplhdr(skb);
2926 u8 credits = hdr->credits;
2927 unsigned int tid = GET_TID(hdr);
2930 ep = get_ep_from_tid(dev, tid);
2933 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2935 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2936 __func__, ep, ep->hwtid, state_read(&ep->com));
2940 dst_confirm(ep->dst);
2942 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2943 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2944 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2945 mutex_lock(&ep->com.mutex);
2946 kfree_skb(ep->mpa_skb);
2948 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2950 mutex_unlock(&ep->com.mutex);
2953 c4iw_put_ep(&ep->com);
2957 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2960 struct c4iw_ep *ep = to_ep(cm_id);
2962 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2964 mutex_lock(&ep->com.mutex);
2965 if (ep->com.state != MPA_REQ_RCVD) {
2966 mutex_unlock(&ep->com.mutex);
2967 c4iw_put_ep(&ep->com);
2970 set_bit(ULP_REJECT, &ep->com.history);
2974 abort = send_mpa_reject(ep, pdata, pdata_len);
2975 mutex_unlock(&ep->com.mutex);
2978 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
2979 c4iw_put_ep(&ep->com);
2983 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2986 struct c4iw_qp_attributes attrs;
2987 enum c4iw_qp_attr_mask mask;
2988 struct c4iw_ep *ep = to_ep(cm_id);
2989 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2990 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2993 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2995 mutex_lock(&ep->com.mutex);
2996 if (ep->com.state != MPA_REQ_RCVD) {
3003 set_bit(ULP_ACCEPT, &ep->com.history);
3004 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3005 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3010 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3011 if (conn_param->ord > ep->ird) {
3012 if (RELAXED_IRD_NEGOTIATION) {
3013 conn_param->ord = ep->ird;
3015 ep->ird = conn_param->ird;
3016 ep->ord = conn_param->ord;
3017 send_mpa_reject(ep, conn_param->private_data,
3018 conn_param->private_data_len);
3023 if (conn_param->ird < ep->ord) {
3024 if (RELAXED_IRD_NEGOTIATION &&
3025 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3026 conn_param->ird = ep->ord;
3033 ep->ird = conn_param->ird;
3034 ep->ord = conn_param->ord;
3036 if (ep->mpa_attr.version == 1) {
3037 if (peer2peer && ep->ird == 0)
3041 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3042 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3046 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3048 ep->com.cm_id = cm_id;
3049 ref_cm_id(&ep->com);
3053 /* bind QP to EP and move to RTS */
3054 attrs.mpa_attr = ep->mpa_attr;
3055 attrs.max_ird = ep->ird;
3056 attrs.max_ord = ep->ord;
3057 attrs.llp_stream_handle = ep;
3058 attrs.next_state = C4IW_QP_STATE_RTS;
3060 /* bind QP and TID with INIT_WR */
3061 mask = C4IW_QP_ATTR_NEXT_STATE |
3062 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3063 C4IW_QP_ATTR_MPA_ATTR |
3064 C4IW_QP_ATTR_MAX_IRD |
3065 C4IW_QP_ATTR_MAX_ORD;
3067 err = c4iw_modify_qp(ep->com.qp->rhp,
3068 ep->com.qp, mask, &attrs, 1);
3070 goto err_deref_cm_id;
3072 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3073 err = send_mpa_reply(ep, conn_param->private_data,
3074 conn_param->private_data_len);
3076 goto err_deref_cm_id;
3078 __state_set(&ep->com, FPDU_MODE);
3079 established_upcall(ep);
3080 mutex_unlock(&ep->com.mutex);
3081 c4iw_put_ep(&ep->com);
3084 deref_cm_id(&ep->com);
3088 mutex_unlock(&ep->com.mutex);
3090 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3091 c4iw_put_ep(&ep->com);
3095 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3097 struct in_device *ind;
3099 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3100 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3102 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3104 return -EADDRNOTAVAIL;
3105 for_primary_ifa(ind) {
3106 laddr->sin_addr.s_addr = ifa->ifa_address;
3107 raddr->sin_addr.s_addr = ifa->ifa_address;
3113 return found ? 0 : -EADDRNOTAVAIL;
3116 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3117 unsigned char banned_flags)
3119 struct inet6_dev *idev;
3120 int err = -EADDRNOTAVAIL;
3123 idev = __in6_dev_get(dev);
3125 struct inet6_ifaddr *ifp;
3127 read_lock_bh(&idev->lock);
3128 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3129 if (ifp->scope == IFA_LINK &&
3130 !(ifp->flags & banned_flags)) {
3131 memcpy(addr, &ifp->addr, 16);
3136 read_unlock_bh(&idev->lock);
3142 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3144 struct in6_addr uninitialized_var(addr);
3145 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3146 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3148 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3149 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3150 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3153 return -EADDRNOTAVAIL;
3156 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3158 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3161 struct sockaddr_in *laddr;
3162 struct sockaddr_in *raddr;
3163 struct sockaddr_in6 *laddr6;
3164 struct sockaddr_in6 *raddr6;
3168 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3169 (conn_param->ird > cur_max_read_depth(dev))) {
3173 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3175 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3180 skb_queue_head_init(&ep->com.ep_skb_list);
3181 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3186 init_timer(&ep->timer);
3187 ep->plen = conn_param->private_data_len;
3189 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3190 conn_param->private_data, ep->plen);
3191 ep->ird = conn_param->ird;
3192 ep->ord = conn_param->ord;
3194 if (peer2peer && ep->ord == 0)
3197 ep->com.cm_id = cm_id;
3198 ref_cm_id(&ep->com);
3200 ep->com.qp = get_qhp(dev, conn_param->qpn);
3202 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3207 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3211 * Allocate an active TID to initiate a TCP connection.
3213 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3214 if (ep->atid == -1) {
3215 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3219 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3221 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3222 sizeof(ep->com.local_addr));
3223 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3224 sizeof(ep->com.remote_addr));
3226 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3227 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3228 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3229 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3231 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3233 ra = (__u8 *)&raddr->sin_addr;
3236 * Handle loopback requests to INADDR_ANY.
3238 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3239 err = pick_local_ipaddrs(dev, cm_id);
3245 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3246 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3247 ra, ntohs(raddr->sin_port));
3248 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3249 laddr->sin_addr.s_addr,
3250 raddr->sin_addr.s_addr,
3252 raddr->sin_port, cm_id->tos);
3255 ra = (__u8 *)&raddr6->sin6_addr;
3258 * Handle loopback requests to INADDR_ANY.
3260 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3261 err = pick_local_ip6addrs(dev, cm_id);
3267 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3268 __func__, laddr6->sin6_addr.s6_addr,
3269 ntohs(laddr6->sin6_port),
3270 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3271 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3272 laddr6->sin6_addr.s6_addr,
3273 raddr6->sin6_addr.s6_addr,
3275 raddr6->sin6_port, 0,
3276 raddr6->sin6_scope_id);
3279 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3280 err = -EHOSTUNREACH;
3284 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3285 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3287 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3291 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3292 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3295 state_set(&ep->com, CONNECTING);
3296 ep->tos = cm_id->tos;
3298 /* send connect request to rnic */
3299 err = send_connect(ep);
3303 cxgb4_l2t_release(ep->l2t);
3305 dst_release(ep->dst);
3307 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3308 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3310 skb_queue_purge(&ep->com.ep_skb_list);
3311 deref_cm_id(&ep->com);
3313 c4iw_put_ep(&ep->com);
3318 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3321 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3322 &ep->com.local_addr;
3324 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3325 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3326 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3330 c4iw_init_wr_wait(&ep->com.wr_wait);
3331 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3332 ep->stid, &sin6->sin6_addr,
3334 ep->com.dev->rdev.lldi.rxq_ids[0]);
3336 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3340 err = net_xmit_errno(err);
3342 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3343 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3344 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3346 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3351 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3354 struct sockaddr_in *sin = (struct sockaddr_in *)
3355 &ep->com.local_addr;
3357 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3359 err = cxgb4_create_server_filter(
3360 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3361 sin->sin_addr.s_addr, sin->sin_port, 0,
3362 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3363 if (err == -EBUSY) {
3364 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3368 set_current_state(TASK_UNINTERRUPTIBLE);
3369 schedule_timeout(usecs_to_jiffies(100));
3371 } while (err == -EBUSY);
3373 c4iw_init_wr_wait(&ep->com.wr_wait);
3374 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3375 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3376 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3378 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3382 err = net_xmit_errno(err);
3385 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3387 &sin->sin_addr, ntohs(sin->sin_port));
3391 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3394 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3395 struct c4iw_listen_ep *ep;
3399 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3401 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3405 skb_queue_head_init(&ep->com.ep_skb_list);
3406 PDBG("%s ep %p\n", __func__, ep);
3407 ep->com.cm_id = cm_id;
3408 ref_cm_id(&ep->com);
3410 ep->backlog = backlog;
3411 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3412 sizeof(ep->com.local_addr));
3415 * Allocate a server TID.
3417 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3418 ep->com.local_addr.ss_family == AF_INET)
3419 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3420 cm_id->m_local_addr.ss_family, ep);
3422 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3423 cm_id->m_local_addr.ss_family, ep);
3425 if (ep->stid == -1) {
3426 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3430 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3432 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3433 sizeof(ep->com.local_addr));
3435 state_set(&ep->com, LISTEN);
3436 if (ep->com.local_addr.ss_family == AF_INET)
3437 err = create_server4(dev, ep);
3439 err = create_server6(dev, ep);
3441 cm_id->provider_data = ep;
3445 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3446 ep->com.local_addr.ss_family);
3448 deref_cm_id(&ep->com);
3449 c4iw_put_ep(&ep->com);
3455 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3458 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3460 PDBG("%s ep %p\n", __func__, ep);
3463 state_set(&ep->com, DEAD);
3464 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3465 ep->com.local_addr.ss_family == AF_INET) {
3466 err = cxgb4_remove_server_filter(
3467 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3468 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3470 struct sockaddr_in6 *sin6;
3471 c4iw_init_wr_wait(&ep->com.wr_wait);
3472 err = cxgb4_remove_server(
3473 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3474 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3477 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3479 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3480 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3481 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3483 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3484 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3485 ep->com.local_addr.ss_family);
3487 deref_cm_id(&ep->com);
3488 c4iw_put_ep(&ep->com);
3492 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3497 struct c4iw_rdev *rdev;
3499 mutex_lock(&ep->com.mutex);
3501 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3502 states[ep->com.state], abrupt);
3505 * Ref the ep here in case we have fatal errors causing the
3506 * ep to be released and freed.
3508 c4iw_get_ep(&ep->com);
3510 rdev = &ep->com.dev->rdev;
3511 if (c4iw_fatal_error(rdev)) {
3513 close_complete_upcall(ep, -EIO);
3514 ep->com.state = DEAD;
3516 switch (ep->com.state) {
3525 ep->com.state = ABORTING;
3527 ep->com.state = CLOSING;
3530 * if we close before we see the fw4_ack() then we fix
3531 * up the timer state since we're reusing it.
3534 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3535 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3540 set_bit(CLOSE_SENT, &ep->com.flags);
3543 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3546 (void)stop_ep_timer(ep);
3547 ep->com.state = ABORTING;
3549 ep->com.state = MORIBUND;
3555 PDBG("%s ignoring disconnect ep %p state %u\n",
3556 __func__, ep, ep->com.state);
3565 set_bit(EP_DISC_ABORT, &ep->com.history);
3566 close_complete_upcall(ep, -ECONNRESET);
3567 ret = send_abort(ep);
3569 set_bit(EP_DISC_CLOSE, &ep->com.history);
3570 ret = send_halfclose(ep);
3573 set_bit(EP_DISC_FAIL, &ep->com.history);
3576 close_complete_upcall(ep, -EIO);
3579 struct c4iw_qp_attributes attrs;
3581 attrs.next_state = C4IW_QP_STATE_ERROR;
3582 ret = c4iw_modify_qp(ep->com.qp->rhp,
3584 C4IW_QP_ATTR_NEXT_STATE,
3588 "%s - qp <- error failed!\n",
3594 mutex_unlock(&ep->com.mutex);
3595 c4iw_put_ep(&ep->com);
3597 release_ep_resources(ep);
3601 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3602 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3605 int atid = be32_to_cpu(req->tid);
3607 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3608 (__force u32) req->tid);
3612 switch (req->retval) {
3614 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3615 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3616 send_fw_act_open_req(ep, atid);
3620 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3621 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3622 send_fw_act_open_req(ep, atid);
3627 pr_info("%s unexpected ofld conn wr retval %d\n",
3628 __func__, req->retval);
3631 pr_err("active ofld_connect_wr failure %d atid %d\n",
3633 mutex_lock(&dev->rdev.stats.lock);
3634 dev->rdev.stats.act_ofld_conn_fails++;
3635 mutex_unlock(&dev->rdev.stats.lock);
3636 connect_reply_upcall(ep, status2errno(req->retval));
3637 state_set(&ep->com, DEAD);
3638 if (ep->com.remote_addr.ss_family == AF_INET6) {
3639 struct sockaddr_in6 *sin6 =
3640 (struct sockaddr_in6 *)&ep->com.local_addr;
3641 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3642 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3644 remove_handle(dev, &dev->atid_idr, atid);
3645 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3646 dst_release(ep->dst);
3647 cxgb4_l2t_release(ep->l2t);
3648 c4iw_put_ep(&ep->com);
3651 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3652 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3654 struct sk_buff *rpl_skb;
3655 struct cpl_pass_accept_req *cpl;
3658 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3661 PDBG("%s passive open failure %d\n", __func__, req->retval);
3662 mutex_lock(&dev->rdev.stats.lock);
3663 dev->rdev.stats.pas_ofld_conn_fails++;
3664 mutex_unlock(&dev->rdev.stats.lock);
3667 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3668 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3669 (__force u32) htonl(
3670 (__force u32) req->tid)));
3671 ret = pass_accept_req(dev, rpl_skb);
3678 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3680 struct cpl_fw6_msg *rpl = cplhdr(skb);
3681 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3683 switch (rpl->type) {
3685 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3687 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3688 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3689 switch (req->t_state) {
3691 active_ofld_conn_reply(dev, skb, req);
3694 passive_ofld_conn_reply(dev, skb, req);
3697 pr_err("%s unexpected ofld conn wr state %d\n",
3698 __func__, req->t_state);
3706 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3709 __be16 hdr_len, vlantag, len;
3711 int tcp_hdr_len, ip_hdr_len;
3713 struct cpl_rx_pkt *cpl = cplhdr(skb);
3714 struct cpl_pass_accept_req *req;
3715 struct tcp_options_received tmp_opt;
3716 struct c4iw_dev *dev;
3717 enum chip_type type;
3719 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3720 /* Store values from cpl_rx_pkt in temporary location. */
3721 vlantag = cpl->vlan;
3723 l2info = cpl->l2info;
3724 hdr_len = cpl->hdr_len;
3727 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3730 * We need to parse the TCP options from SYN packet.
3731 * to generate cpl_pass_accept_req.
3733 memset(&tmp_opt, 0, sizeof(tmp_opt));
3734 tcp_clear_options(&tmp_opt);
3735 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3737 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3738 memset(req, 0, sizeof(*req));
3739 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3740 SYN_MAC_IDX_V(RX_MACIDX_G(
3741 be32_to_cpu(l2info))) |
3743 type = dev->rdev.lldi.adapter_type;
3744 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3745 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3747 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3748 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3749 eth_hdr_len = is_t4(type) ?
3750 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3751 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3752 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3753 IP_HDR_LEN_V(ip_hdr_len) |
3754 ETH_HDR_LEN_V(eth_hdr_len));
3755 } else { /* T6 and later */
3756 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3757 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3758 T6_IP_HDR_LEN_V(ip_hdr_len) |
3759 T6_ETH_HDR_LEN_V(eth_hdr_len));
3761 req->vlan = vlantag;
3763 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3764 PASS_OPEN_TOS_V(tos));
3765 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3766 if (tmp_opt.wscale_ok)
3767 req->tcpopt.wsf = tmp_opt.snd_wscale;
3768 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3769 if (tmp_opt.sack_ok)
3770 req->tcpopt.sack = 1;
3771 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3775 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3776 __be32 laddr, __be16 lport,
3777 __be32 raddr, __be16 rport,
3778 u32 rcv_isn, u32 filter, u16 window,
3779 u32 rss_qid, u8 port_id)
3781 struct sk_buff *req_skb;
3782 struct fw_ofld_connection_wr *req;
3783 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3786 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3787 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3788 memset(req, 0, sizeof(*req));
3789 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3790 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3791 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3792 req->le.filter = (__force __be32) filter;
3793 req->le.lport = lport;
3794 req->le.pport = rport;
3795 req->le.u.ipv4.lip = laddr;
3796 req->le.u.ipv4.pip = raddr;
3797 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3798 req->tcb.rcv_adv = htons(window);
3799 req->tcb.t_state_to_astid =
3800 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3801 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3802 FW_OFLD_CONNECTION_WR_ASTID_V(
3803 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3806 * We store the qid in opt2 which will be used by the firmware
3807 * to send us the wr response.
3809 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3812 * We initialize the MSS index in TCB to 0xF.
3813 * So that when driver sends cpl_pass_accept_rpl
3814 * TCB picks up the correct value. If this was 0
3815 * TP will ignore any value > 0 for MSS index.
3817 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3818 req->cookie = (uintptr_t)skb;
3820 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3821 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3823 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3831 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3832 * messages when a filter is being used instead of server to
3833 * redirect a syn packet. When packets hit filter they are redirected
3834 * to the offload queue and driver tries to establish the connection
3835 * using firmware work request.
3837 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3840 unsigned int filter;
3841 struct ethhdr *eh = NULL;
3842 struct vlan_ethhdr *vlan_eh = NULL;
3844 struct tcphdr *tcph;
3845 struct rss_header *rss = (void *)skb->data;
3846 struct cpl_rx_pkt *cpl = (void *)skb->data;
3847 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3848 struct l2t_entry *e;
3849 struct dst_entry *dst;
3850 struct c4iw_ep *lep = NULL;
3852 struct port_info *pi;
3853 struct net_device *pdev;
3854 u16 rss_qid, eth_hdr_len;
3857 struct neighbour *neigh;
3859 /* Drop all non-SYN packets */
3860 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3864 * Drop all packets which did not hit the filter.
3865 * Unlikely to happen.
3867 if (!(rss->filter_hit && rss->filter_tid))
3871 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3873 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3875 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3877 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3881 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3883 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3886 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3889 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3892 pr_err("T%d Chip is not supported\n",
3893 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3897 if (eth_hdr_len == ETH_HLEN) {
3898 eh = (struct ethhdr *)(req + 1);
3899 iph = (struct iphdr *)(eh + 1);
3901 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3902 iph = (struct iphdr *)(vlan_eh + 1);
3903 skb->vlan_tci = ntohs(cpl->vlan);
3906 if (iph->version != 0x4)
3909 tcph = (struct tcphdr *)(iph + 1);
3910 skb_set_network_header(skb, (void *)iph - (void *)rss);
3911 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3914 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3915 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3916 ntohs(tcph->source), iph->tos);
3918 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3919 iph->daddr, iph->saddr, tcph->dest,
3920 tcph->source, iph->tos);
3922 pr_err("%s - failed to find dst entry!\n",
3926 neigh = dst_neigh_lookup_skb(dst, skb);
3929 pr_err("%s - failed to allocate neigh!\n",
3934 if (neigh->dev->flags & IFF_LOOPBACK) {
3935 pdev = ip_dev_find(&init_net, iph->daddr);
3936 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3938 pi = (struct port_info *)netdev_priv(pdev);
3939 tx_chan = cxgb4_port_chan(pdev);
3942 pdev = get_real_dev(neigh->dev);
3943 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3945 pi = (struct port_info *)netdev_priv(pdev);
3946 tx_chan = cxgb4_port_chan(pdev);
3948 neigh_release(neigh);
3950 pr_err("%s - failed to allocate l2t entry!\n",
3955 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3956 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3957 window = (__force u16) htons((__force u16)tcph->window);
3959 /* Calcuate filter portion for LE region. */
3960 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3961 dev->rdev.lldi.ports[0],
3965 * Synthesize the cpl_pass_accept_req. We have everything except the
3966 * TID. Once firmware sends a reply with TID we update the TID field
3967 * in cpl and pass it through the regular cpl_pass_accept_req path.
3969 build_cpl_pass_accept_req(skb, stid, iph->tos);
3970 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3971 tcph->source, ntohl(tcph->seq), filter, window,
3972 rss_qid, pi->port_id);
3973 cxgb4_l2t_release(e);
3978 c4iw_put_ep(&lep->com);
3983 * These are the real handlers that are called from a
3986 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
3987 [CPL_ACT_ESTABLISH] = act_establish,
3988 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3989 [CPL_RX_DATA] = rx_data,
3990 [CPL_ABORT_RPL_RSS] = abort_rpl,
3991 [CPL_ABORT_RPL] = abort_rpl,
3992 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3993 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3994 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3995 [CPL_PASS_ESTABLISH] = pass_establish,
3996 [CPL_PEER_CLOSE] = peer_close,
3997 [CPL_ABORT_REQ_RSS] = peer_abort,
3998 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3999 [CPL_RDMA_TERMINATE] = terminate,
4000 [CPL_FW4_ACK] = fw4_ack,
4001 [CPL_FW6_MSG] = deferred_fw6_msg,
4002 [CPL_RX_PKT] = rx_pkt,
4003 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4004 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4007 static void process_timeout(struct c4iw_ep *ep)
4009 struct c4iw_qp_attributes attrs;
4012 mutex_lock(&ep->com.mutex);
4013 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4015 set_bit(TIMEDOUT, &ep->com.history);
4016 switch (ep->com.state) {
4018 connect_reply_upcall(ep, -ETIMEDOUT);
4027 if (ep->com.cm_id && ep->com.qp) {
4028 attrs.next_state = C4IW_QP_STATE_ERROR;
4029 c4iw_modify_qp(ep->com.qp->rhp,
4030 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4033 close_complete_upcall(ep, -ETIMEDOUT);
4039 * These states are expected if the ep timed out at the same
4040 * time as another thread was calling stop_ep_timer().
4041 * So we silently do nothing for these states.
4046 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4047 __func__, ep, ep->hwtid, ep->com.state);
4050 mutex_unlock(&ep->com.mutex);
4052 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4053 c4iw_put_ep(&ep->com);
4056 static void process_timedout_eps(void)
4060 spin_lock_irq(&timeout_lock);
4061 while (!list_empty(&timeout_list)) {
4062 struct list_head *tmp;
4064 tmp = timeout_list.next;
4068 spin_unlock_irq(&timeout_lock);
4069 ep = list_entry(tmp, struct c4iw_ep, entry);
4070 process_timeout(ep);
4071 spin_lock_irq(&timeout_lock);
4073 spin_unlock_irq(&timeout_lock);
4076 static void process_work(struct work_struct *work)
4078 struct sk_buff *skb = NULL;
4079 struct c4iw_dev *dev;
4080 struct cpl_act_establish *rpl;
4081 unsigned int opcode;
4084 process_timedout_eps();
4085 while ((skb = skb_dequeue(&rxq))) {
4087 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4088 opcode = rpl->ot.opcode;
4090 BUG_ON(!work_handlers[opcode]);
4091 ret = work_handlers[opcode](dev, skb);
4094 process_timedout_eps();
4098 static DECLARE_WORK(skb_work, process_work);
4100 static void ep_timeout(unsigned long arg)
4102 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4105 spin_lock(&timeout_lock);
4106 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4108 * Only insert if it is not already on the list.
4110 if (!ep->entry.next) {
4111 list_add_tail(&ep->entry, &timeout_list);
4115 spin_unlock(&timeout_lock);
4117 queue_work(workq, &skb_work);
4121 * All the CM events are handled on a work queue to have a safe context.
4123 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4127 * Save dev in the skb->cb area.
4129 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4132 * Queue the skb and schedule the worker thread.
4134 skb_queue_tail(&rxq, skb);
4135 queue_work(workq, &skb_work);
4139 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4141 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4143 if (rpl->status != CPL_ERR_NONE) {
4144 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4145 "for tid %u\n", rpl->status, GET_TID(rpl));
4151 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4153 struct cpl_fw6_msg *rpl = cplhdr(skb);
4154 struct c4iw_wr_wait *wr_waitp;
4157 PDBG("%s type %u\n", __func__, rpl->type);
4159 switch (rpl->type) {
4160 case FW6_TYPE_WR_RPL:
4161 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4162 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4163 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4165 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4169 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4173 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4181 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4183 struct cpl_abort_req_rss *req = cplhdr(skb);
4185 unsigned int tid = GET_TID(req);
4187 ep = get_ep_from_tid(dev, tid);
4188 /* This EP will be dereferenced in peer_abort() */
4190 printk(KERN_WARNING MOD
4191 "Abort on non-existent endpoint, tid %d\n", tid);
4195 if (cxgb_is_neg_adv(req->status)) {
4196 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4197 __func__, ep->hwtid, req->status,
4198 neg_adv_str(req->status));
4201 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4204 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4211 * Most upcalls from the T4 Core go to sched() to
4212 * schedule the processing on a work queue.
4214 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4215 [CPL_ACT_ESTABLISH] = sched,
4216 [CPL_ACT_OPEN_RPL] = sched,
4217 [CPL_RX_DATA] = sched,
4218 [CPL_ABORT_RPL_RSS] = sched,
4219 [CPL_ABORT_RPL] = sched,
4220 [CPL_PASS_OPEN_RPL] = sched,
4221 [CPL_CLOSE_LISTSRV_RPL] = sched,
4222 [CPL_PASS_ACCEPT_REQ] = sched,
4223 [CPL_PASS_ESTABLISH] = sched,
4224 [CPL_PEER_CLOSE] = sched,
4225 [CPL_CLOSE_CON_RPL] = sched,
4226 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4227 [CPL_RDMA_TERMINATE] = sched,
4228 [CPL_FW4_ACK] = sched,
4229 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4230 [CPL_FW6_MSG] = fw6_msg,
4231 [CPL_RX_PKT] = sched
4234 int __init c4iw_cm_init(void)
4236 spin_lock_init(&timeout_lock);
4237 skb_queue_head_init(&rxq);
4239 workq = create_singlethread_workqueue("iw_cxgb4");
4246 void c4iw_cm_term(void)
4248 WARN_ON(!list_empty(&timeout_list));
4249 flush_workqueue(workq);
4250 destroy_workqueue(workq);