2 * Copyright (c) 2006 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/slab.h>
35 #include <linux/workqueue.h>
36 #include <linux/skbuff.h>
37 #include <linux/timer.h>
38 #include <linux/notifier.h>
39 #include <linux/inetdevice.h>
41 #include <net/neighbour.h>
42 #include <net/netevent.h>
43 #include <net/route.h>
46 #include "cxgb3_offload.h"
48 #include "iwch_provider.h"
51 static char *states[] = {
68 module_param(peer2peer, int, 0644);
69 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
71 static int ep_timeout_secs = 60;
72 module_param(ep_timeout_secs, int, 0644);
73 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
74 "in seconds (default=60)");
76 static int mpa_rev = 1;
77 module_param(mpa_rev, int, 0644);
78 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
79 "1 is spec compliant. (default=1)");
81 static int markers_enabled = 0;
82 module_param(markers_enabled, int, 0644);
83 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
85 static int crc_enabled = 1;
86 module_param(crc_enabled, int, 0644);
87 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
89 static int rcv_win = 256 * 1024;
90 module_param(rcv_win, int, 0644);
91 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
93 static int snd_win = 32 * 1024;
94 module_param(snd_win, int, 0644);
95 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
97 static unsigned int nocong = 0;
98 module_param(nocong, uint, 0644);
99 MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
101 static unsigned int cong_flavor = 1;
102 module_param(cong_flavor, uint, 0644);
103 MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
105 static struct workqueue_struct *workq;
107 static struct sk_buff_head rxq;
109 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
110 static void ep_timeout(unsigned long arg);
111 static void connect_reply_upcall(struct iwch_ep *ep, int status);
113 static void start_ep_timer(struct iwch_ep *ep)
115 PDBG("%s ep %p\n", __func__, ep);
116 if (timer_pending(&ep->timer)) {
117 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
118 del_timer_sync(&ep->timer);
121 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
122 ep->timer.data = (unsigned long)ep;
123 ep->timer.function = ep_timeout;
124 add_timer(&ep->timer);
127 static void stop_ep_timer(struct iwch_ep *ep)
129 PDBG("%s ep %p\n", __func__, ep);
130 if (!timer_pending(&ep->timer)) {
131 printk(KERN_ERR "%s timer stopped when its not running! ep %p state %u\n",
132 __func__, ep, ep->com.state);
136 del_timer_sync(&ep->timer);
140 static int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
143 struct cxio_rdev *rdev;
145 rdev = (struct cxio_rdev *)tdev->ulp;
146 if (cxio_fatal_error(rdev)) {
150 error = l2t_send(tdev, skb, l2e);
156 int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
159 struct cxio_rdev *rdev;
161 rdev = (struct cxio_rdev *)tdev->ulp;
162 if (cxio_fatal_error(rdev)) {
166 error = cxgb3_ofld_send(tdev, skb);
172 static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
174 struct cpl_tid_release *req;
176 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
179 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
180 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
181 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
182 skb->priority = CPL_PRIORITY_SETUP;
183 iwch_cxgb3_ofld_send(tdev, skb);
187 int iwch_quiesce_tid(struct iwch_ep *ep)
189 struct cpl_set_tcb_field *req;
190 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
194 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
195 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
196 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
197 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
200 req->word = htons(W_TCB_RX_QUIESCE);
201 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
202 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
204 skb->priority = CPL_PRIORITY_DATA;
205 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
208 int iwch_resume_tid(struct iwch_ep *ep)
210 struct cpl_set_tcb_field *req;
211 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
215 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
216 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
217 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
218 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
221 req->word = htons(W_TCB_RX_QUIESCE);
222 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
225 skb->priority = CPL_PRIORITY_DATA;
226 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
229 static void set_emss(struct iwch_ep *ep, u16 opt)
231 PDBG("%s ep %p opt %u\n", __func__, ep, opt);
232 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
233 if (G_TCPOPT_TSTAMP(opt))
237 PDBG("emss=%d\n", ep->emss);
240 static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
243 enum iwch_ep_state state;
245 spin_lock_irqsave(&epc->lock, flags);
247 spin_unlock_irqrestore(&epc->lock, flags);
251 static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
256 static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
260 spin_lock_irqsave(&epc->lock, flags);
261 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
262 __state_set(epc, new);
263 spin_unlock_irqrestore(&epc->lock, flags);
267 static void *alloc_ep(int size, gfp_t gfp)
269 struct iwch_ep_common *epc;
271 epc = kzalloc(size, gfp);
273 kref_init(&epc->kref);
274 spin_lock_init(&epc->lock);
275 init_waitqueue_head(&epc->waitq);
277 PDBG("%s alloc ep %p\n", __func__, epc);
281 void __free_ep(struct kref *kref)
284 ep = container_of(container_of(kref, struct iwch_ep_common, kref),
285 struct iwch_ep, com);
286 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
287 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
288 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
289 dst_release(ep->dst);
290 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
295 static void release_ep_resources(struct iwch_ep *ep)
297 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
298 set_bit(RELEASE_RESOURCES, &ep->com.flags);
302 static int status2errno(int status)
307 case CPL_ERR_CONN_RESET:
309 case CPL_ERR_ARP_MISS:
310 return -EHOSTUNREACH;
311 case CPL_ERR_CONN_TIMEDOUT:
313 case CPL_ERR_TCAM_FULL:
315 case CPL_ERR_CONN_EXIST:
323 * Try and reuse skbs already allocated...
325 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
327 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
331 skb = alloc_skb(len, gfp);
336 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
337 __be32 peer_ip, __be16 local_port,
338 __be16 peer_port, u8 tos)
349 .proto = IPPROTO_TCP,
357 rt = ip_route_output_flow(&init_net, &fl, NULL);
363 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
367 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
372 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
374 PDBG("%s t3cdev %p\n", __func__, dev);
379 * Handle an ARP failure for an active open.
381 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
383 printk(KERN_ERR MOD "ARP failure duing connect\n");
388 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
391 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
393 struct cpl_abort_req *req = cplhdr(skb);
395 PDBG("%s t3cdev %p\n", __func__, dev);
396 req->cmd = CPL_ABORT_NO_RST;
397 iwch_cxgb3_ofld_send(dev, skb);
400 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
402 struct cpl_close_con_req *req;
405 PDBG("%s ep %p\n", __func__, ep);
406 skb = get_skb(NULL, sizeof(*req), gfp);
408 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
411 skb->priority = CPL_PRIORITY_DATA;
412 set_arp_failure_handler(skb, arp_failure_discard);
413 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
414 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
415 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
416 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
417 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
420 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
422 struct cpl_abort_req *req;
424 PDBG("%s ep %p\n", __func__, ep);
425 skb = get_skb(skb, sizeof(*req), gfp);
427 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
431 skb->priority = CPL_PRIORITY_DATA;
432 set_arp_failure_handler(skb, abort_arp_failure);
433 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
434 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
435 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
436 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
437 req->cmd = CPL_ABORT_SEND_RST;
438 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
441 static int send_connect(struct iwch_ep *ep)
443 struct cpl_act_open_req *req;
445 u32 opt0h, opt0l, opt2;
446 unsigned int mtu_idx;
449 PDBG("%s ep %p\n", __func__, ep);
451 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
453 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
457 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
458 wscale = compute_wscale(rcv_win);
463 V_WND_SCALE(wscale) |
465 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
466 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
467 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
468 V_CONG_CONTROL_FLAVOR(cong_flavor);
469 skb->priority = CPL_PRIORITY_SETUP;
470 set_arp_failure_handler(skb, act_open_req_arp_failure);
472 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
473 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
474 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
475 req->local_port = ep->com.local_addr.sin_port;
476 req->peer_port = ep->com.remote_addr.sin_port;
477 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
478 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
479 req->opt0h = htonl(opt0h);
480 req->opt0l = htonl(opt0l);
482 req->opt2 = htonl(opt2);
483 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
486 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
489 struct tx_data_wr *req;
490 struct mpa_message *mpa;
493 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
495 BUG_ON(skb_cloned(skb));
497 mpalen = sizeof(*mpa) + ep->plen;
498 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
500 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
502 connect_reply_upcall(ep, -ENOMEM);
507 skb_reserve(skb, sizeof(*req));
508 skb_put(skb, mpalen);
509 skb->priority = CPL_PRIORITY_DATA;
510 mpa = (struct mpa_message *) skb->data;
511 memset(mpa, 0, sizeof(*mpa));
512 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
513 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
514 (markers_enabled ? MPA_MARKERS : 0);
515 mpa->private_data_size = htons(ep->plen);
516 mpa->revision = mpa_rev;
519 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
522 * Reference the mpa skb. This ensures the data area
523 * will remain in memory until the hw acks the tx.
524 * Function tx_ack() will deref it.
527 set_arp_failure_handler(skb, arp_failure_discard);
528 skb_reset_transport_header(skb);
530 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
531 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
532 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
533 req->len = htonl(len);
534 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
535 V_TX_SNDBUF(snd_win>>15));
536 req->flags = htonl(F_TX_INIT);
537 req->sndseq = htonl(ep->snd_seq);
540 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
542 state_set(&ep->com, MPA_REQ_SENT);
546 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
549 struct tx_data_wr *req;
550 struct mpa_message *mpa;
553 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
555 mpalen = sizeof(*mpa) + plen;
557 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
559 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
562 skb_reserve(skb, sizeof(*req));
563 mpa = (struct mpa_message *) skb_put(skb, mpalen);
564 memset(mpa, 0, sizeof(*mpa));
565 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
566 mpa->flags = MPA_REJECT;
567 mpa->revision = mpa_rev;
568 mpa->private_data_size = htons(plen);
570 memcpy(mpa->private_data, pdata, plen);
573 * Reference the mpa skb again. This ensures the data area
574 * will remain in memory until the hw acks the tx.
575 * Function tx_ack() will deref it.
578 skb->priority = CPL_PRIORITY_DATA;
579 set_arp_failure_handler(skb, arp_failure_discard);
580 skb_reset_transport_header(skb);
581 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
582 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
583 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
584 req->len = htonl(mpalen);
585 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
586 V_TX_SNDBUF(snd_win>>15));
587 req->flags = htonl(F_TX_INIT);
588 req->sndseq = htonl(ep->snd_seq);
591 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
594 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
597 struct tx_data_wr *req;
598 struct mpa_message *mpa;
602 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
604 mpalen = sizeof(*mpa) + plen;
606 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
608 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
611 skb->priority = CPL_PRIORITY_DATA;
612 skb_reserve(skb, sizeof(*req));
613 mpa = (struct mpa_message *) skb_put(skb, mpalen);
614 memset(mpa, 0, sizeof(*mpa));
615 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
616 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
617 (markers_enabled ? MPA_MARKERS : 0);
618 mpa->revision = mpa_rev;
619 mpa->private_data_size = htons(plen);
621 memcpy(mpa->private_data, pdata, plen);
624 * Reference the mpa skb. This ensures the data area
625 * will remain in memory until the hw acks the tx.
626 * Function tx_ack() will deref it.
629 set_arp_failure_handler(skb, arp_failure_discard);
630 skb_reset_transport_header(skb);
632 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
633 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
634 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
635 req->len = htonl(len);
636 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
637 V_TX_SNDBUF(snd_win>>15));
638 req->flags = htonl(F_TX_INIT);
639 req->sndseq = htonl(ep->snd_seq);
641 state_set(&ep->com, MPA_REP_SENT);
642 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
645 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
647 struct iwch_ep *ep = ctx;
648 struct cpl_act_establish *req = cplhdr(skb);
649 unsigned int tid = GET_TID(req);
651 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
653 dst_confirm(ep->dst);
655 /* setup the hwtid for this connection */
657 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
659 ep->snd_seq = ntohl(req->snd_isn);
660 ep->rcv_seq = ntohl(req->rcv_isn);
662 set_emss(ep, ntohs(req->tcp_opt));
664 /* dealloc the atid */
665 cxgb3_free_atid(ep->com.tdev, ep->atid);
667 /* start MPA negotiation */
668 send_mpa_req(ep, skb);
673 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
675 PDBG("%s ep %p\n", __FILE__, ep);
676 state_set(&ep->com, ABORTING);
677 send_abort(ep, skb, gfp);
680 static void close_complete_upcall(struct iwch_ep *ep)
682 struct iw_cm_event event;
684 PDBG("%s ep %p\n", __func__, ep);
685 memset(&event, 0, sizeof(event));
686 event.event = IW_CM_EVENT_CLOSE;
688 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
689 ep, ep->com.cm_id, ep->hwtid);
690 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
691 ep->com.cm_id->rem_ref(ep->com.cm_id);
692 ep->com.cm_id = NULL;
697 static void peer_close_upcall(struct iwch_ep *ep)
699 struct iw_cm_event event;
701 PDBG("%s ep %p\n", __func__, ep);
702 memset(&event, 0, sizeof(event));
703 event.event = IW_CM_EVENT_DISCONNECT;
705 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
706 ep, ep->com.cm_id, ep->hwtid);
707 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
711 static void peer_abort_upcall(struct iwch_ep *ep)
713 struct iw_cm_event event;
715 PDBG("%s ep %p\n", __func__, ep);
716 memset(&event, 0, sizeof(event));
717 event.event = IW_CM_EVENT_CLOSE;
718 event.status = -ECONNRESET;
720 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
721 ep->com.cm_id, ep->hwtid);
722 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
723 ep->com.cm_id->rem_ref(ep->com.cm_id);
724 ep->com.cm_id = NULL;
729 static void connect_reply_upcall(struct iwch_ep *ep, int status)
731 struct iw_cm_event event;
733 PDBG("%s ep %p status %d\n", __func__, ep, status);
734 memset(&event, 0, sizeof(event));
735 event.event = IW_CM_EVENT_CONNECT_REPLY;
736 event.status = status;
737 event.local_addr = ep->com.local_addr;
738 event.remote_addr = ep->com.remote_addr;
740 if ((status == 0) || (status == -ECONNREFUSED)) {
741 event.private_data_len = ep->plen;
742 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
745 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
747 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
750 ep->com.cm_id->rem_ref(ep->com.cm_id);
751 ep->com.cm_id = NULL;
756 static void connect_request_upcall(struct iwch_ep *ep)
758 struct iw_cm_event event;
760 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
761 memset(&event, 0, sizeof(event));
762 event.event = IW_CM_EVENT_CONNECT_REQUEST;
763 event.local_addr = ep->com.local_addr;
764 event.remote_addr = ep->com.remote_addr;
765 event.private_data_len = ep->plen;
766 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
767 event.provider_data = ep;
768 if (state_read(&ep->parent_ep->com) != DEAD) {
770 ep->parent_ep->com.cm_id->event_handler(
771 ep->parent_ep->com.cm_id,
774 put_ep(&ep->parent_ep->com);
775 ep->parent_ep = NULL;
778 static void established_upcall(struct iwch_ep *ep)
780 struct iw_cm_event event;
782 PDBG("%s ep %p\n", __func__, ep);
783 memset(&event, 0, sizeof(event));
784 event.event = IW_CM_EVENT_ESTABLISHED;
786 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
787 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
791 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
793 struct cpl_rx_data_ack *req;
796 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
797 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
799 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
803 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
804 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
805 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
806 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
807 skb->priority = CPL_PRIORITY_ACK;
808 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
812 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
814 struct mpa_message *mpa;
816 struct iwch_qp_attributes attrs;
817 enum iwch_qp_attr_mask mask;
820 PDBG("%s ep %p\n", __func__, ep);
823 * Stop mpa timer. If it expired, then the state has
824 * changed and we bail since ep_timeout already aborted
828 if (state_read(&ep->com) != MPA_REQ_SENT)
832 * If we get more than the supported amount of private data
833 * then we must fail this connection.
835 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
841 * copy the new data into our accumulation buffer.
843 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
845 ep->mpa_pkt_len += skb->len;
848 * if we don't even have the mpa message, then bail.
850 if (ep->mpa_pkt_len < sizeof(*mpa))
852 mpa = (struct mpa_message *) ep->mpa_pkt;
854 /* Validate MPA header. */
855 if (mpa->revision != mpa_rev) {
859 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
864 plen = ntohs(mpa->private_data_size);
867 * Fail if there's too much private data.
869 if (plen > MPA_MAX_PRIVATE_DATA) {
875 * If plen does not account for pkt size
877 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
882 ep->plen = (u8) plen;
885 * If we don't have all the pdata yet, then bail.
886 * We'll continue process when more data arrives.
888 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
891 if (mpa->flags & MPA_REJECT) {
897 * If we get here we have accumulated the entire mpa
898 * start reply message including private data. And
899 * the MPA header is valid.
901 state_set(&ep->com, FPDU_MODE);
902 ep->mpa_attr.initiator = 1;
903 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
904 ep->mpa_attr.recv_marker_enabled = markers_enabled;
905 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
906 ep->mpa_attr.version = mpa_rev;
907 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
908 "xmit_marker_enabled=%d, version=%d\n", __func__,
909 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
910 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
912 attrs.mpa_attr = ep->mpa_attr;
913 attrs.max_ird = ep->ird;
914 attrs.max_ord = ep->ord;
915 attrs.llp_stream_handle = ep;
916 attrs.next_state = IWCH_QP_STATE_RTS;
918 mask = IWCH_QP_ATTR_NEXT_STATE |
919 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
920 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
922 /* bind QP and TID with INIT_WR */
923 err = iwch_modify_qp(ep->com.qp->rhp,
924 ep->com.qp, mask, &attrs, 1);
928 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
929 iwch_post_zb_read(ep->com.qp);
934 abort_connection(ep, skb, GFP_KERNEL);
936 connect_reply_upcall(ep, err);
940 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
942 struct mpa_message *mpa;
945 PDBG("%s ep %p\n", __func__, ep);
948 * Stop mpa timer. If it expired, then the state has
949 * changed and we bail since ep_timeout already aborted
953 if (state_read(&ep->com) != MPA_REQ_WAIT)
957 * If we get more than the supported amount of private data
958 * then we must fail this connection.
960 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
961 abort_connection(ep, skb, GFP_KERNEL);
965 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
968 * Copy the new data into our accumulation buffer.
970 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
972 ep->mpa_pkt_len += skb->len;
975 * If we don't even have the mpa message, then bail.
976 * We'll continue process when more data arrives.
978 if (ep->mpa_pkt_len < sizeof(*mpa))
980 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
981 mpa = (struct mpa_message *) ep->mpa_pkt;
984 * Validate MPA Header.
986 if (mpa->revision != mpa_rev) {
987 abort_connection(ep, skb, GFP_KERNEL);
991 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
992 abort_connection(ep, skb, GFP_KERNEL);
996 plen = ntohs(mpa->private_data_size);
999 * Fail if there's too much private data.
1001 if (plen > MPA_MAX_PRIVATE_DATA) {
1002 abort_connection(ep, skb, GFP_KERNEL);
1007 * If plen does not account for pkt size
1009 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1010 abort_connection(ep, skb, GFP_KERNEL);
1013 ep->plen = (u8) plen;
1016 * If we don't have all the pdata yet, then bail.
1018 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1022 * If we get here we have accumulated the entire mpa
1023 * start reply message including private data.
1025 ep->mpa_attr.initiator = 0;
1026 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1027 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1028 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1029 ep->mpa_attr.version = mpa_rev;
1030 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1031 "xmit_marker_enabled=%d, version=%d\n", __func__,
1032 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1033 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1035 state_set(&ep->com, MPA_REQ_RCVD);
1038 connect_request_upcall(ep);
1042 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1044 struct iwch_ep *ep = ctx;
1045 struct cpl_rx_data *hdr = cplhdr(skb);
1046 unsigned int dlen = ntohs(hdr->len);
1048 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1050 skb_pull(skb, sizeof(*hdr));
1051 skb_trim(skb, dlen);
1053 ep->rcv_seq += dlen;
1054 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1056 switch (state_read(&ep->com)) {
1058 process_mpa_reply(ep, skb);
1061 process_mpa_request(ep, skb);
1066 printk(KERN_ERR MOD "%s Unexpected streaming data."
1067 " ep %p state %d tid %d\n",
1068 __func__, ep, state_read(&ep->com), ep->hwtid);
1071 * The ep will timeout and inform the ULP of the failure.
1077 /* update RX credits */
1078 update_rx_credits(ep, dlen);
1080 return CPL_RET_BUF_DONE;
1084 * Upcall from the adapter indicating data has been transmitted.
1085 * For us its just the single MPA request or reply. We can now free
1086 * the skb holding the mpa message.
1088 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1090 struct iwch_ep *ep = ctx;
1091 struct cpl_wr_ack *hdr = cplhdr(skb);
1092 unsigned int credits = ntohs(hdr->credits);
1094 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1097 PDBG("%s 0 credit ack ep %p state %u\n",
1098 __func__, ep, state_read(&ep->com));
1099 return CPL_RET_BUF_DONE;
1102 BUG_ON(credits != 1);
1103 dst_confirm(ep->dst);
1105 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1106 __func__, ep, state_read(&ep->com));
1107 if (ep->mpa_attr.initiator) {
1108 PDBG("%s initiator ep %p state %u\n",
1109 __func__, ep, state_read(&ep->com));
1111 iwch_post_zb_read(ep->com.qp);
1113 PDBG("%s responder ep %p state %u\n",
1114 __func__, ep, state_read(&ep->com));
1115 ep->com.rpl_done = 1;
1116 wake_up(&ep->com.waitq);
1119 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1120 __func__, ep, state_read(&ep->com));
1121 kfree_skb(ep->mpa_skb);
1124 return CPL_RET_BUF_DONE;
1127 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1129 struct iwch_ep *ep = ctx;
1130 unsigned long flags;
1133 PDBG("%s ep %p\n", __func__, ep);
1137 * We get 2 abort replies from the HW. The first one must
1138 * be ignored except for scribbling that we need one more.
1140 if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1141 return CPL_RET_BUF_DONE;
1144 spin_lock_irqsave(&ep->com.lock, flags);
1145 switch (ep->com.state) {
1147 close_complete_upcall(ep);
1148 __state_set(&ep->com, DEAD);
1152 printk(KERN_ERR "%s ep %p state %d\n",
1153 __func__, ep, ep->com.state);
1156 spin_unlock_irqrestore(&ep->com.lock, flags);
1159 release_ep_resources(ep);
1160 return CPL_RET_BUF_DONE;
1164 * Return whether a failed active open has allocated a TID
1166 static inline int act_open_has_tid(int status)
1168 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1169 status != CPL_ERR_ARP_MISS;
1172 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1174 struct iwch_ep *ep = ctx;
1175 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1177 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1178 status2errno(rpl->status));
1179 connect_reply_upcall(ep, status2errno(rpl->status));
1180 state_set(&ep->com, DEAD);
1181 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1182 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1183 cxgb3_free_atid(ep->com.tdev, ep->atid);
1184 dst_release(ep->dst);
1185 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1187 return CPL_RET_BUF_DONE;
1190 static int listen_start(struct iwch_listen_ep *ep)
1192 struct sk_buff *skb;
1193 struct cpl_pass_open_req *req;
1195 PDBG("%s ep %p\n", __func__, ep);
1196 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1198 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1202 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1203 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1204 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1205 req->local_port = ep->com.local_addr.sin_port;
1206 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1209 req->peer_netmask = 0;
1210 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1211 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1212 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1215 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1218 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1220 struct iwch_listen_ep *ep = ctx;
1221 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1223 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1224 rpl->status, status2errno(rpl->status));
1225 ep->com.rpl_err = status2errno(rpl->status);
1226 ep->com.rpl_done = 1;
1227 wake_up(&ep->com.waitq);
1229 return CPL_RET_BUF_DONE;
1232 static int listen_stop(struct iwch_listen_ep *ep)
1234 struct sk_buff *skb;
1235 struct cpl_close_listserv_req *req;
1237 PDBG("%s ep %p\n", __func__, ep);
1238 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1240 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1243 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1244 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1246 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1248 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1251 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1254 struct iwch_listen_ep *ep = ctx;
1255 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1257 PDBG("%s ep %p\n", __func__, ep);
1258 ep->com.rpl_err = status2errno(rpl->status);
1259 ep->com.rpl_done = 1;
1260 wake_up(&ep->com.waitq);
1261 return CPL_RET_BUF_DONE;
1264 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1266 struct cpl_pass_accept_rpl *rpl;
1267 unsigned int mtu_idx;
1268 u32 opt0h, opt0l, opt2;
1271 PDBG("%s ep %p\n", __func__, ep);
1272 BUG_ON(skb_cloned(skb));
1273 skb_trim(skb, sizeof(*rpl));
1275 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1276 wscale = compute_wscale(rcv_win);
1277 opt0h = V_NAGLE(0) |
1281 V_WND_SCALE(wscale) |
1282 V_MSS_IDX(mtu_idx) |
1283 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1284 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1285 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1286 V_CONG_CONTROL_FLAVOR(cong_flavor);
1289 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1290 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1291 rpl->peer_ip = peer_ip;
1292 rpl->opt0h = htonl(opt0h);
1293 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1294 rpl->opt2 = htonl(opt2);
1295 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1296 skb->priority = CPL_PRIORITY_SETUP;
1297 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1302 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1303 struct sk_buff *skb)
1305 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1307 BUG_ON(skb_cloned(skb));
1308 skb_trim(skb, sizeof(struct cpl_tid_release));
1311 if (tdev->type != T3A)
1312 release_tid(tdev, hwtid, skb);
1314 struct cpl_pass_accept_rpl *rpl;
1317 skb->priority = CPL_PRIORITY_SETUP;
1318 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1319 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1321 rpl->peer_ip = peer_ip;
1322 rpl->opt0h = htonl(F_TCAM_BYPASS);
1323 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1325 rpl->rsvd = rpl->opt2;
1326 iwch_cxgb3_ofld_send(tdev, skb);
1330 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1332 struct iwch_ep *child_ep, *parent_ep = ctx;
1333 struct cpl_pass_accept_req *req = cplhdr(skb);
1334 unsigned int hwtid = GET_TID(req);
1335 struct dst_entry *dst;
1336 struct l2t_entry *l2t;
1340 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1342 if (state_read(&parent_ep->com) != LISTEN) {
1343 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1349 * Find the netdev for this connection request.
1351 tim.mac_addr = req->dst_mac;
1352 tim.vlan_tag = ntohs(req->vlan_tag);
1353 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1354 printk(KERN_ERR "%s bad dst mac %pM\n",
1355 __func__, req->dst_mac);
1359 /* Find output route */
1360 rt = find_route(tdev,
1364 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1366 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1371 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1373 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1378 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1380 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1382 l2t_release(L2DATA(tdev), l2t);
1386 state_set(&child_ep->com, CONNECTING);
1387 child_ep->com.tdev = tdev;
1388 child_ep->com.cm_id = NULL;
1389 child_ep->com.local_addr.sin_family = PF_INET;
1390 child_ep->com.local_addr.sin_port = req->local_port;
1391 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1392 child_ep->com.remote_addr.sin_family = PF_INET;
1393 child_ep->com.remote_addr.sin_port = req->peer_port;
1394 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1395 get_ep(&parent_ep->com);
1396 child_ep->parent_ep = parent_ep;
1397 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1398 child_ep->l2t = l2t;
1399 child_ep->dst = dst;
1400 child_ep->hwtid = hwtid;
1401 init_timer(&child_ep->timer);
1402 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1403 accept_cr(child_ep, req->peer_ip, skb);
1406 reject_cr(tdev, hwtid, req->peer_ip, skb);
1408 return CPL_RET_BUF_DONE;
1411 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1413 struct iwch_ep *ep = ctx;
1414 struct cpl_pass_establish *req = cplhdr(skb);
1416 PDBG("%s ep %p\n", __func__, ep);
1417 ep->snd_seq = ntohl(req->snd_isn);
1418 ep->rcv_seq = ntohl(req->rcv_isn);
1420 set_emss(ep, ntohs(req->tcp_opt));
1422 dst_confirm(ep->dst);
1423 state_set(&ep->com, MPA_REQ_WAIT);
1426 return CPL_RET_BUF_DONE;
1429 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1431 struct iwch_ep *ep = ctx;
1432 struct iwch_qp_attributes attrs;
1433 unsigned long flags;
1437 PDBG("%s ep %p\n", __func__, ep);
1438 dst_confirm(ep->dst);
1440 spin_lock_irqsave(&ep->com.lock, flags);
1441 switch (ep->com.state) {
1443 __state_set(&ep->com, CLOSING);
1446 __state_set(&ep->com, CLOSING);
1447 connect_reply_upcall(ep, -ECONNRESET);
1452 * We're gonna mark this puppy DEAD, but keep
1453 * the reference on it until the ULP accepts or
1454 * rejects the CR. Also wake up anyone waiting
1455 * in rdma connection migration (see iwch_accept_cr()).
1457 __state_set(&ep->com, CLOSING);
1458 ep->com.rpl_done = 1;
1459 ep->com.rpl_err = -ECONNRESET;
1460 PDBG("waking up ep %p\n", ep);
1461 wake_up(&ep->com.waitq);
1464 __state_set(&ep->com, CLOSING);
1465 ep->com.rpl_done = 1;
1466 ep->com.rpl_err = -ECONNRESET;
1467 PDBG("waking up ep %p\n", ep);
1468 wake_up(&ep->com.waitq);
1472 __state_set(&ep->com, CLOSING);
1473 attrs.next_state = IWCH_QP_STATE_CLOSING;
1474 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1475 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1476 peer_close_upcall(ep);
1482 __state_set(&ep->com, MORIBUND);
1487 if (ep->com.cm_id && ep->com.qp) {
1488 attrs.next_state = IWCH_QP_STATE_IDLE;
1489 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1490 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1492 close_complete_upcall(ep);
1493 __state_set(&ep->com, DEAD);
1503 spin_unlock_irqrestore(&ep->com.lock, flags);
1505 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1507 release_ep_resources(ep);
1508 return CPL_RET_BUF_DONE;
1512 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1514 static int is_neg_adv_abort(unsigned int status)
1516 return status == CPL_ERR_RTX_NEG_ADVICE ||
1517 status == CPL_ERR_PERSIST_NEG_ADVICE;
1520 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1522 struct cpl_abort_req_rss *req = cplhdr(skb);
1523 struct iwch_ep *ep = ctx;
1524 struct cpl_abort_rpl *rpl;
1525 struct sk_buff *rpl_skb;
1526 struct iwch_qp_attributes attrs;
1529 unsigned long flags;
1531 if (is_neg_adv_abort(req->status)) {
1532 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1534 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1535 return CPL_RET_BUF_DONE;
1539 * We get 2 peer aborts from the HW. The first one must
1540 * be ignored except for scribbling that we need one more.
1542 if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
1543 return CPL_RET_BUF_DONE;
1546 spin_lock_irqsave(&ep->com.lock, flags);
1547 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1548 switch (ep->com.state) {
1556 connect_reply_upcall(ep, -ECONNRESET);
1559 ep->com.rpl_done = 1;
1560 ep->com.rpl_err = -ECONNRESET;
1561 PDBG("waking up ep %p\n", ep);
1562 wake_up(&ep->com.waitq);
1567 * We're gonna mark this puppy DEAD, but keep
1568 * the reference on it until the ULP accepts or
1569 * rejects the CR. Also wake up anyone waiting
1570 * in rdma connection migration (see iwch_accept_cr()).
1572 ep->com.rpl_done = 1;
1573 ep->com.rpl_err = -ECONNRESET;
1574 PDBG("waking up ep %p\n", ep);
1575 wake_up(&ep->com.waitq);
1582 if (ep->com.cm_id && ep->com.qp) {
1583 attrs.next_state = IWCH_QP_STATE_ERROR;
1584 ret = iwch_modify_qp(ep->com.qp->rhp,
1585 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1589 "%s - qp <- error failed!\n",
1592 peer_abort_upcall(ep);
1597 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1598 spin_unlock_irqrestore(&ep->com.lock, flags);
1599 return CPL_RET_BUF_DONE;
1604 dst_confirm(ep->dst);
1605 if (ep->com.state != ABORTING) {
1606 __state_set(&ep->com, DEAD);
1609 spin_unlock_irqrestore(&ep->com.lock, flags);
1611 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1613 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1618 rpl_skb->priority = CPL_PRIORITY_DATA;
1619 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1620 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1621 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1622 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1623 rpl->cmd = CPL_ABORT_NO_RST;
1624 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1627 release_ep_resources(ep);
1628 return CPL_RET_BUF_DONE;
1631 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1633 struct iwch_ep *ep = ctx;
1634 struct iwch_qp_attributes attrs;
1635 unsigned long flags;
1638 PDBG("%s ep %p\n", __func__, ep);
1641 /* The cm_id may be null if we failed to connect */
1642 spin_lock_irqsave(&ep->com.lock, flags);
1643 switch (ep->com.state) {
1645 __state_set(&ep->com, MORIBUND);
1649 if ((ep->com.cm_id) && (ep->com.qp)) {
1650 attrs.next_state = IWCH_QP_STATE_IDLE;
1651 iwch_modify_qp(ep->com.qp->rhp,
1653 IWCH_QP_ATTR_NEXT_STATE,
1656 close_complete_upcall(ep);
1657 __state_set(&ep->com, DEAD);
1667 spin_unlock_irqrestore(&ep->com.lock, flags);
1669 release_ep_resources(ep);
1670 return CPL_RET_BUF_DONE;
1674 * T3A does 3 things when a TERM is received:
1675 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1676 * 2) generate an async event on the QP with the TERMINATE opcode
1677 * 3) post a TERMINATE opcde cqe into the associated CQ.
1679 * For (1), we save the message in the qp for later consumer consumption.
1680 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1681 * For (3), we toss the CQE in cxio_poll_cq().
1683 * terminate() handles case (1)...
1685 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1687 struct iwch_ep *ep = ctx;
1689 if (state_read(&ep->com) != FPDU_MODE)
1690 return CPL_RET_BUF_DONE;
1692 PDBG("%s ep %p\n", __func__, ep);
1693 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1694 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1695 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1697 ep->com.qp->attr.terminate_msg_len = skb->len;
1698 ep->com.qp->attr.is_terminate_local = 0;
1699 return CPL_RET_BUF_DONE;
1702 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1704 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1705 struct iwch_ep *ep = ctx;
1707 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
1710 struct iwch_qp_attributes attrs;
1712 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1713 __func__, ep->hwtid);
1715 attrs.next_state = IWCH_QP_STATE_ERROR;
1716 iwch_modify_qp(ep->com.qp->rhp,
1717 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1719 abort_connection(ep, NULL, GFP_KERNEL);
1721 return CPL_RET_BUF_DONE;
1724 static void ep_timeout(unsigned long arg)
1726 struct iwch_ep *ep = (struct iwch_ep *)arg;
1727 struct iwch_qp_attributes attrs;
1728 unsigned long flags;
1731 spin_lock_irqsave(&ep->com.lock, flags);
1732 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
1734 switch (ep->com.state) {
1736 __state_set(&ep->com, ABORTING);
1737 connect_reply_upcall(ep, -ETIMEDOUT);
1740 __state_set(&ep->com, ABORTING);
1744 if (ep->com.cm_id && ep->com.qp) {
1745 attrs.next_state = IWCH_QP_STATE_ERROR;
1746 iwch_modify_qp(ep->com.qp->rhp,
1747 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1750 __state_set(&ep->com, ABORTING);
1753 printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1754 __func__, ep, ep->com.state);
1758 spin_unlock_irqrestore(&ep->com.lock, flags);
1760 abort_connection(ep, NULL, GFP_ATOMIC);
1764 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1767 struct iwch_ep *ep = to_ep(cm_id);
1768 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1770 if (state_read(&ep->com) == DEAD) {
1774 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1776 abort_connection(ep, NULL, GFP_KERNEL);
1778 err = send_mpa_reject(ep, pdata, pdata_len);
1779 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1785 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1788 struct iwch_qp_attributes attrs;
1789 enum iwch_qp_attr_mask mask;
1790 struct iwch_ep *ep = to_ep(cm_id);
1791 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1792 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1794 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1795 if (state_read(&ep->com) == DEAD) {
1800 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1803 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1804 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1805 abort_connection(ep, NULL, GFP_KERNEL);
1810 cm_id->add_ref(cm_id);
1811 ep->com.cm_id = cm_id;
1814 ep->ird = conn_param->ird;
1815 ep->ord = conn_param->ord;
1817 if (peer2peer && ep->ird == 0)
1820 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1822 /* bind QP to EP and move to RTS */
1823 attrs.mpa_attr = ep->mpa_attr;
1824 attrs.max_ird = ep->ird;
1825 attrs.max_ord = ep->ord;
1826 attrs.llp_stream_handle = ep;
1827 attrs.next_state = IWCH_QP_STATE_RTS;
1829 /* bind QP and TID with INIT_WR */
1830 mask = IWCH_QP_ATTR_NEXT_STATE |
1831 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1832 IWCH_QP_ATTR_MPA_ATTR |
1833 IWCH_QP_ATTR_MAX_IRD |
1834 IWCH_QP_ATTR_MAX_ORD;
1836 err = iwch_modify_qp(ep->com.qp->rhp,
1837 ep->com.qp, mask, &attrs, 1);
1841 /* if needed, wait for wr_ack */
1842 if (iwch_rqes_posted(qp)) {
1843 wait_event(ep->com.waitq, ep->com.rpl_done);
1844 err = ep->com.rpl_err;
1849 err = send_mpa_reply(ep, conn_param->private_data,
1850 conn_param->private_data_len);
1855 state_set(&ep->com, FPDU_MODE);
1856 established_upcall(ep);
1860 ep->com.cm_id = NULL;
1862 cm_id->rem_ref(cm_id);
1868 static int is_loopback_dst(struct iw_cm_id *cm_id)
1870 struct net_device *dev;
1872 dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1879 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1882 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1886 if (is_loopback_dst(cm_id)) {
1891 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1893 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1897 init_timer(&ep->timer);
1898 ep->plen = conn_param->private_data_len;
1900 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1901 conn_param->private_data, ep->plen);
1902 ep->ird = conn_param->ird;
1903 ep->ord = conn_param->ord;
1905 if (peer2peer && ep->ord == 0)
1908 ep->com.tdev = h->rdev.t3cdev_p;
1910 cm_id->add_ref(cm_id);
1911 ep->com.cm_id = cm_id;
1912 ep->com.qp = get_qhp(h, conn_param->qpn);
1913 BUG_ON(!ep->com.qp);
1914 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1918 * Allocate an active TID to initiate a TCP connection.
1920 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1921 if (ep->atid == -1) {
1922 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1928 rt = find_route(h->rdev.t3cdev_p,
1929 cm_id->local_addr.sin_addr.s_addr,
1930 cm_id->remote_addr.sin_addr.s_addr,
1931 cm_id->local_addr.sin_port,
1932 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1934 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1935 err = -EHOSTUNREACH;
1940 /* get a l2t entry */
1941 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1942 ep->dst->neighbour->dev);
1944 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1949 state_set(&ep->com, CONNECTING);
1950 ep->tos = IPTOS_LOWDELAY;
1951 ep->com.local_addr = cm_id->local_addr;
1952 ep->com.remote_addr = cm_id->remote_addr;
1954 /* send connect request to rnic */
1955 err = send_connect(ep);
1959 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1961 dst_release(ep->dst);
1963 cxgb3_free_atid(ep->com.tdev, ep->atid);
1965 cm_id->rem_ref(cm_id);
1971 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1974 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1975 struct iwch_listen_ep *ep;
1980 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1982 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1986 PDBG("%s ep %p\n", __func__, ep);
1987 ep->com.tdev = h->rdev.t3cdev_p;
1988 cm_id->add_ref(cm_id);
1989 ep->com.cm_id = cm_id;
1990 ep->backlog = backlog;
1991 ep->com.local_addr = cm_id->local_addr;
1994 * Allocate a server TID.
1996 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1997 if (ep->stid == -1) {
1998 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2003 state_set(&ep->com, LISTEN);
2004 err = listen_start(ep);
2008 /* wait for pass_open_rpl */
2009 wait_event(ep->com.waitq, ep->com.rpl_done);
2010 err = ep->com.rpl_err;
2012 cm_id->provider_data = ep;
2016 cxgb3_free_stid(ep->com.tdev, ep->stid);
2018 cm_id->rem_ref(cm_id);
2025 int iwch_destroy_listen(struct iw_cm_id *cm_id)
2028 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2030 PDBG("%s ep %p\n", __func__, ep);
2033 state_set(&ep->com, DEAD);
2034 ep->com.rpl_done = 0;
2035 ep->com.rpl_err = 0;
2036 err = listen_stop(ep);
2039 wait_event(ep->com.waitq, ep->com.rpl_done);
2040 cxgb3_free_stid(ep->com.tdev, ep->stid);
2042 err = ep->com.rpl_err;
2043 cm_id->rem_ref(cm_id);
2048 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2051 unsigned long flags;
2054 struct t3cdev *tdev;
2055 struct cxio_rdev *rdev;
2057 spin_lock_irqsave(&ep->com.lock, flags);
2059 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2060 states[ep->com.state], abrupt);
2062 tdev = (struct t3cdev *)ep->com.tdev;
2063 rdev = (struct cxio_rdev *)tdev->ulp;
2064 if (cxio_fatal_error(rdev)) {
2066 close_complete_upcall(ep);
2067 ep->com.state = DEAD;
2069 switch (ep->com.state) {
2077 ep->com.state = ABORTING;
2079 ep->com.state = CLOSING;
2082 set_bit(CLOSE_SENT, &ep->com.flags);
2085 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2089 ep->com.state = ABORTING;
2091 ep->com.state = MORIBUND;
2097 PDBG("%s ignoring disconnect ep %p state %u\n",
2098 __func__, ep, ep->com.state);
2105 spin_unlock_irqrestore(&ep->com.lock, flags);
2108 ret = send_abort(ep, NULL, gfp);
2110 ret = send_halfclose(ep, gfp);
2115 release_ep_resources(ep);
2119 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2120 struct l2t_entry *l2t)
2122 struct iwch_ep *ep = ctx;
2127 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2130 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2138 * All the CM events are handled on a work queue to have a safe context.
2139 * These are the real handlers that are called from the work queue.
2141 static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2142 [CPL_ACT_ESTABLISH] = act_establish,
2143 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2144 [CPL_RX_DATA] = rx_data,
2145 [CPL_TX_DMA_ACK] = tx_ack,
2146 [CPL_ABORT_RPL_RSS] = abort_rpl,
2147 [CPL_ABORT_RPL] = abort_rpl,
2148 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2149 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2150 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2151 [CPL_PASS_ESTABLISH] = pass_establish,
2152 [CPL_PEER_CLOSE] = peer_close,
2153 [CPL_ABORT_REQ_RSS] = peer_abort,
2154 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2155 [CPL_RDMA_TERMINATE] = terminate,
2156 [CPL_RDMA_EC_STATUS] = ec_status,
2159 static void process_work(struct work_struct *work)
2161 struct sk_buff *skb = NULL;
2163 struct t3cdev *tdev;
2166 while ((skb = skb_dequeue(&rxq))) {
2167 ep = *((void **) (skb->cb));
2168 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2169 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2170 if (ret & CPL_RET_BUF_DONE)
2174 * ep was referenced in sched(), and is freed here.
2176 put_ep((struct iwch_ep_common *)ep);
2180 static DECLARE_WORK(skb_work, process_work);
2182 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2184 struct iwch_ep_common *epc = ctx;
2189 * Save ctx and tdev in the skb->cb area.
2191 *((void **) skb->cb) = ctx;
2192 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2195 * Queue the skb and schedule the worker thread.
2197 skb_queue_tail(&rxq, skb);
2198 queue_work(workq, &skb_work);
2202 static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2204 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2206 if (rpl->status != CPL_ERR_NONE) {
2207 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2208 "for tid %u\n", rpl->status, GET_TID(rpl));
2210 return CPL_RET_BUF_DONE;
2214 * All upcalls from the T3 Core go to sched() to schedule the
2215 * processing on a work queue.
2217 cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2218 [CPL_ACT_ESTABLISH] = sched,
2219 [CPL_ACT_OPEN_RPL] = sched,
2220 [CPL_RX_DATA] = sched,
2221 [CPL_TX_DMA_ACK] = sched,
2222 [CPL_ABORT_RPL_RSS] = sched,
2223 [CPL_ABORT_RPL] = sched,
2224 [CPL_PASS_OPEN_RPL] = sched,
2225 [CPL_CLOSE_LISTSRV_RPL] = sched,
2226 [CPL_PASS_ACCEPT_REQ] = sched,
2227 [CPL_PASS_ESTABLISH] = sched,
2228 [CPL_PEER_CLOSE] = sched,
2229 [CPL_CLOSE_CON_RPL] = sched,
2230 [CPL_ABORT_REQ_RSS] = sched,
2231 [CPL_RDMA_TERMINATE] = sched,
2232 [CPL_RDMA_EC_STATUS] = sched,
2233 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2236 int __init iwch_cm_init(void)
2238 skb_queue_head_init(&rxq);
2240 workq = create_singlethread_workqueue("iw_cxgb3");
2247 void __exit iwch_cm_term(void)
2249 flush_workqueue(workq);
2250 destroy_workqueue(workq);