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)
342 rt = ip_route_output_ports(&init_net, NULL, peer_ip, local_ip,
343 peer_port, local_port, IPPROTO_TCP,
350 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
354 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
359 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
361 PDBG("%s t3cdev %p\n", __func__, dev);
366 * Handle an ARP failure for an active open.
368 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
370 printk(KERN_ERR MOD "ARP failure duing connect\n");
375 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
378 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
380 struct cpl_abort_req *req = cplhdr(skb);
382 PDBG("%s t3cdev %p\n", __func__, dev);
383 req->cmd = CPL_ABORT_NO_RST;
384 iwch_cxgb3_ofld_send(dev, skb);
387 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
389 struct cpl_close_con_req *req;
392 PDBG("%s ep %p\n", __func__, ep);
393 skb = get_skb(NULL, sizeof(*req), gfp);
395 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
398 skb->priority = CPL_PRIORITY_DATA;
399 set_arp_failure_handler(skb, arp_failure_discard);
400 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
401 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
402 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
403 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
404 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
407 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
409 struct cpl_abort_req *req;
411 PDBG("%s ep %p\n", __func__, ep);
412 skb = get_skb(skb, sizeof(*req), gfp);
414 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
418 skb->priority = CPL_PRIORITY_DATA;
419 set_arp_failure_handler(skb, abort_arp_failure);
420 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
421 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
422 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
423 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
424 req->cmd = CPL_ABORT_SEND_RST;
425 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
428 static int send_connect(struct iwch_ep *ep)
430 struct cpl_act_open_req *req;
432 u32 opt0h, opt0l, opt2;
433 unsigned int mtu_idx;
436 PDBG("%s ep %p\n", __func__, ep);
438 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
440 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
444 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
445 wscale = compute_wscale(rcv_win);
450 V_WND_SCALE(wscale) |
452 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
453 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
454 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
455 V_CONG_CONTROL_FLAVOR(cong_flavor);
456 skb->priority = CPL_PRIORITY_SETUP;
457 set_arp_failure_handler(skb, act_open_req_arp_failure);
459 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
460 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
461 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
462 req->local_port = ep->com.local_addr.sin_port;
463 req->peer_port = ep->com.remote_addr.sin_port;
464 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
465 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
466 req->opt0h = htonl(opt0h);
467 req->opt0l = htonl(opt0l);
469 req->opt2 = htonl(opt2);
470 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
473 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
476 struct tx_data_wr *req;
477 struct mpa_message *mpa;
480 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
482 BUG_ON(skb_cloned(skb));
484 mpalen = sizeof(*mpa) + ep->plen;
485 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
487 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
489 connect_reply_upcall(ep, -ENOMEM);
494 skb_reserve(skb, sizeof(*req));
495 skb_put(skb, mpalen);
496 skb->priority = CPL_PRIORITY_DATA;
497 mpa = (struct mpa_message *) skb->data;
498 memset(mpa, 0, sizeof(*mpa));
499 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
500 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
501 (markers_enabled ? MPA_MARKERS : 0);
502 mpa->private_data_size = htons(ep->plen);
503 mpa->revision = mpa_rev;
506 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
509 * Reference the mpa skb. This ensures the data area
510 * will remain in memory until the hw acks the tx.
511 * Function tx_ack() will deref it.
514 set_arp_failure_handler(skb, arp_failure_discard);
515 skb_reset_transport_header(skb);
517 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
518 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
519 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
520 req->len = htonl(len);
521 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
522 V_TX_SNDBUF(snd_win>>15));
523 req->flags = htonl(F_TX_INIT);
524 req->sndseq = htonl(ep->snd_seq);
527 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
529 state_set(&ep->com, MPA_REQ_SENT);
533 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
536 struct tx_data_wr *req;
537 struct mpa_message *mpa;
540 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
542 mpalen = sizeof(*mpa) + plen;
544 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
546 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
549 skb_reserve(skb, sizeof(*req));
550 mpa = (struct mpa_message *) skb_put(skb, mpalen);
551 memset(mpa, 0, sizeof(*mpa));
552 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
553 mpa->flags = MPA_REJECT;
554 mpa->revision = mpa_rev;
555 mpa->private_data_size = htons(plen);
557 memcpy(mpa->private_data, pdata, plen);
560 * Reference the mpa skb again. This ensures the data area
561 * will remain in memory until the hw acks the tx.
562 * Function tx_ack() will deref it.
565 skb->priority = CPL_PRIORITY_DATA;
566 set_arp_failure_handler(skb, arp_failure_discard);
567 skb_reset_transport_header(skb);
568 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
569 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
570 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
571 req->len = htonl(mpalen);
572 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
573 V_TX_SNDBUF(snd_win>>15));
574 req->flags = htonl(F_TX_INIT);
575 req->sndseq = htonl(ep->snd_seq);
578 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
581 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
584 struct tx_data_wr *req;
585 struct mpa_message *mpa;
589 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
591 mpalen = sizeof(*mpa) + plen;
593 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
595 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
598 skb->priority = CPL_PRIORITY_DATA;
599 skb_reserve(skb, sizeof(*req));
600 mpa = (struct mpa_message *) skb_put(skb, mpalen);
601 memset(mpa, 0, sizeof(*mpa));
602 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
603 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
604 (markers_enabled ? MPA_MARKERS : 0);
605 mpa->revision = mpa_rev;
606 mpa->private_data_size = htons(plen);
608 memcpy(mpa->private_data, pdata, plen);
611 * Reference the mpa skb. This ensures the data area
612 * will remain in memory until the hw acks the tx.
613 * Function tx_ack() will deref it.
616 set_arp_failure_handler(skb, arp_failure_discard);
617 skb_reset_transport_header(skb);
619 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
620 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
621 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
622 req->len = htonl(len);
623 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
624 V_TX_SNDBUF(snd_win>>15));
625 req->flags = htonl(F_TX_INIT);
626 req->sndseq = htonl(ep->snd_seq);
628 state_set(&ep->com, MPA_REP_SENT);
629 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
632 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
634 struct iwch_ep *ep = ctx;
635 struct cpl_act_establish *req = cplhdr(skb);
636 unsigned int tid = GET_TID(req);
638 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
640 dst_confirm(ep->dst);
642 /* setup the hwtid for this connection */
644 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
646 ep->snd_seq = ntohl(req->snd_isn);
647 ep->rcv_seq = ntohl(req->rcv_isn);
649 set_emss(ep, ntohs(req->tcp_opt));
651 /* dealloc the atid */
652 cxgb3_free_atid(ep->com.tdev, ep->atid);
654 /* start MPA negotiation */
655 send_mpa_req(ep, skb);
660 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
662 PDBG("%s ep %p\n", __FILE__, ep);
663 state_set(&ep->com, ABORTING);
664 send_abort(ep, skb, gfp);
667 static void close_complete_upcall(struct iwch_ep *ep)
669 struct iw_cm_event event;
671 PDBG("%s ep %p\n", __func__, ep);
672 memset(&event, 0, sizeof(event));
673 event.event = IW_CM_EVENT_CLOSE;
675 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
676 ep, ep->com.cm_id, ep->hwtid);
677 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
678 ep->com.cm_id->rem_ref(ep->com.cm_id);
679 ep->com.cm_id = NULL;
684 static void peer_close_upcall(struct iwch_ep *ep)
686 struct iw_cm_event event;
688 PDBG("%s ep %p\n", __func__, ep);
689 memset(&event, 0, sizeof(event));
690 event.event = IW_CM_EVENT_DISCONNECT;
692 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
693 ep, ep->com.cm_id, ep->hwtid);
694 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
698 static void peer_abort_upcall(struct iwch_ep *ep)
700 struct iw_cm_event event;
702 PDBG("%s ep %p\n", __func__, ep);
703 memset(&event, 0, sizeof(event));
704 event.event = IW_CM_EVENT_CLOSE;
705 event.status = -ECONNRESET;
707 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
708 ep->com.cm_id, ep->hwtid);
709 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
710 ep->com.cm_id->rem_ref(ep->com.cm_id);
711 ep->com.cm_id = NULL;
716 static void connect_reply_upcall(struct iwch_ep *ep, int status)
718 struct iw_cm_event event;
720 PDBG("%s ep %p status %d\n", __func__, ep, status);
721 memset(&event, 0, sizeof(event));
722 event.event = IW_CM_EVENT_CONNECT_REPLY;
723 event.status = status;
724 event.local_addr = ep->com.local_addr;
725 event.remote_addr = ep->com.remote_addr;
727 if ((status == 0) || (status == -ECONNREFUSED)) {
728 event.private_data_len = ep->plen;
729 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
732 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
734 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
737 ep->com.cm_id->rem_ref(ep->com.cm_id);
738 ep->com.cm_id = NULL;
743 static void connect_request_upcall(struct iwch_ep *ep)
745 struct iw_cm_event event;
747 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
748 memset(&event, 0, sizeof(event));
749 event.event = IW_CM_EVENT_CONNECT_REQUEST;
750 event.local_addr = ep->com.local_addr;
751 event.remote_addr = ep->com.remote_addr;
752 event.private_data_len = ep->plen;
753 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
754 event.provider_data = ep;
755 if (state_read(&ep->parent_ep->com) != DEAD) {
757 ep->parent_ep->com.cm_id->event_handler(
758 ep->parent_ep->com.cm_id,
761 put_ep(&ep->parent_ep->com);
762 ep->parent_ep = NULL;
765 static void established_upcall(struct iwch_ep *ep)
767 struct iw_cm_event event;
769 PDBG("%s ep %p\n", __func__, ep);
770 memset(&event, 0, sizeof(event));
771 event.event = IW_CM_EVENT_ESTABLISHED;
773 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
774 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
778 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
780 struct cpl_rx_data_ack *req;
783 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
784 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
786 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
790 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
791 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
792 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
793 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
794 skb->priority = CPL_PRIORITY_ACK;
795 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
799 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
801 struct mpa_message *mpa;
803 struct iwch_qp_attributes attrs;
804 enum iwch_qp_attr_mask mask;
807 PDBG("%s ep %p\n", __func__, ep);
810 * Stop mpa timer. If it expired, then the state has
811 * changed and we bail since ep_timeout already aborted
815 if (state_read(&ep->com) != MPA_REQ_SENT)
819 * If we get more than the supported amount of private data
820 * then we must fail this connection.
822 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
828 * copy the new data into our accumulation buffer.
830 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
832 ep->mpa_pkt_len += skb->len;
835 * if we don't even have the mpa message, then bail.
837 if (ep->mpa_pkt_len < sizeof(*mpa))
839 mpa = (struct mpa_message *) ep->mpa_pkt;
841 /* Validate MPA header. */
842 if (mpa->revision != mpa_rev) {
846 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
851 plen = ntohs(mpa->private_data_size);
854 * Fail if there's too much private data.
856 if (plen > MPA_MAX_PRIVATE_DATA) {
862 * If plen does not account for pkt size
864 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
869 ep->plen = (u8) plen;
872 * If we don't have all the pdata yet, then bail.
873 * We'll continue process when more data arrives.
875 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
878 if (mpa->flags & MPA_REJECT) {
884 * If we get here we have accumulated the entire mpa
885 * start reply message including private data. And
886 * the MPA header is valid.
888 state_set(&ep->com, FPDU_MODE);
889 ep->mpa_attr.initiator = 1;
890 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
891 ep->mpa_attr.recv_marker_enabled = markers_enabled;
892 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
893 ep->mpa_attr.version = mpa_rev;
894 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
895 "xmit_marker_enabled=%d, version=%d\n", __func__,
896 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
897 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
899 attrs.mpa_attr = ep->mpa_attr;
900 attrs.max_ird = ep->ird;
901 attrs.max_ord = ep->ord;
902 attrs.llp_stream_handle = ep;
903 attrs.next_state = IWCH_QP_STATE_RTS;
905 mask = IWCH_QP_ATTR_NEXT_STATE |
906 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
907 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
909 /* bind QP and TID with INIT_WR */
910 err = iwch_modify_qp(ep->com.qp->rhp,
911 ep->com.qp, mask, &attrs, 1);
915 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
916 iwch_post_zb_read(ep->com.qp);
921 abort_connection(ep, skb, GFP_KERNEL);
923 connect_reply_upcall(ep, err);
927 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
929 struct mpa_message *mpa;
932 PDBG("%s ep %p\n", __func__, ep);
935 * Stop mpa timer. If it expired, then the state has
936 * changed and we bail since ep_timeout already aborted
940 if (state_read(&ep->com) != MPA_REQ_WAIT)
944 * If we get more than the supported amount of private data
945 * then we must fail this connection.
947 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
948 abort_connection(ep, skb, GFP_KERNEL);
952 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
955 * Copy the new data into our accumulation buffer.
957 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
959 ep->mpa_pkt_len += skb->len;
962 * If we don't even have the mpa message, then bail.
963 * We'll continue process when more data arrives.
965 if (ep->mpa_pkt_len < sizeof(*mpa))
967 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
968 mpa = (struct mpa_message *) ep->mpa_pkt;
971 * Validate MPA Header.
973 if (mpa->revision != mpa_rev) {
974 abort_connection(ep, skb, GFP_KERNEL);
978 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
979 abort_connection(ep, skb, GFP_KERNEL);
983 plen = ntohs(mpa->private_data_size);
986 * Fail if there's too much private data.
988 if (plen > MPA_MAX_PRIVATE_DATA) {
989 abort_connection(ep, skb, GFP_KERNEL);
994 * If plen does not account for pkt size
996 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
997 abort_connection(ep, skb, GFP_KERNEL);
1000 ep->plen = (u8) plen;
1003 * If we don't have all the pdata yet, then bail.
1005 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1009 * If we get here we have accumulated the entire mpa
1010 * start reply message including private data.
1012 ep->mpa_attr.initiator = 0;
1013 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1014 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1015 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1016 ep->mpa_attr.version = mpa_rev;
1017 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1018 "xmit_marker_enabled=%d, version=%d\n", __func__,
1019 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1020 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1022 state_set(&ep->com, MPA_REQ_RCVD);
1025 connect_request_upcall(ep);
1029 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1031 struct iwch_ep *ep = ctx;
1032 struct cpl_rx_data *hdr = cplhdr(skb);
1033 unsigned int dlen = ntohs(hdr->len);
1035 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1037 skb_pull(skb, sizeof(*hdr));
1038 skb_trim(skb, dlen);
1040 ep->rcv_seq += dlen;
1041 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1043 switch (state_read(&ep->com)) {
1045 process_mpa_reply(ep, skb);
1048 process_mpa_request(ep, skb);
1053 printk(KERN_ERR MOD "%s Unexpected streaming data."
1054 " ep %p state %d tid %d\n",
1055 __func__, ep, state_read(&ep->com), ep->hwtid);
1058 * The ep will timeout and inform the ULP of the failure.
1064 /* update RX credits */
1065 update_rx_credits(ep, dlen);
1067 return CPL_RET_BUF_DONE;
1071 * Upcall from the adapter indicating data has been transmitted.
1072 * For us its just the single MPA request or reply. We can now free
1073 * the skb holding the mpa message.
1075 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1077 struct iwch_ep *ep = ctx;
1078 struct cpl_wr_ack *hdr = cplhdr(skb);
1079 unsigned int credits = ntohs(hdr->credits);
1081 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1084 PDBG("%s 0 credit ack ep %p state %u\n",
1085 __func__, ep, state_read(&ep->com));
1086 return CPL_RET_BUF_DONE;
1089 BUG_ON(credits != 1);
1090 dst_confirm(ep->dst);
1092 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1093 __func__, ep, state_read(&ep->com));
1094 if (ep->mpa_attr.initiator) {
1095 PDBG("%s initiator ep %p state %u\n",
1096 __func__, ep, state_read(&ep->com));
1098 iwch_post_zb_read(ep->com.qp);
1100 PDBG("%s responder ep %p state %u\n",
1101 __func__, ep, state_read(&ep->com));
1102 ep->com.rpl_done = 1;
1103 wake_up(&ep->com.waitq);
1106 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1107 __func__, ep, state_read(&ep->com));
1108 kfree_skb(ep->mpa_skb);
1111 return CPL_RET_BUF_DONE;
1114 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1116 struct iwch_ep *ep = ctx;
1117 unsigned long flags;
1120 PDBG("%s ep %p\n", __func__, ep);
1124 * We get 2 abort replies from the HW. The first one must
1125 * be ignored except for scribbling that we need one more.
1127 if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1128 return CPL_RET_BUF_DONE;
1131 spin_lock_irqsave(&ep->com.lock, flags);
1132 switch (ep->com.state) {
1134 close_complete_upcall(ep);
1135 __state_set(&ep->com, DEAD);
1139 printk(KERN_ERR "%s ep %p state %d\n",
1140 __func__, ep, ep->com.state);
1143 spin_unlock_irqrestore(&ep->com.lock, flags);
1146 release_ep_resources(ep);
1147 return CPL_RET_BUF_DONE;
1151 * Return whether a failed active open has allocated a TID
1153 static inline int act_open_has_tid(int status)
1155 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1156 status != CPL_ERR_ARP_MISS;
1159 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1161 struct iwch_ep *ep = ctx;
1162 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1164 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1165 status2errno(rpl->status));
1166 connect_reply_upcall(ep, status2errno(rpl->status));
1167 state_set(&ep->com, DEAD);
1168 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1169 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1170 cxgb3_free_atid(ep->com.tdev, ep->atid);
1171 dst_release(ep->dst);
1172 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1174 return CPL_RET_BUF_DONE;
1177 static int listen_start(struct iwch_listen_ep *ep)
1179 struct sk_buff *skb;
1180 struct cpl_pass_open_req *req;
1182 PDBG("%s ep %p\n", __func__, ep);
1183 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1185 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1189 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1190 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1191 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1192 req->local_port = ep->com.local_addr.sin_port;
1193 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1196 req->peer_netmask = 0;
1197 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1198 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1199 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1202 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1205 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1207 struct iwch_listen_ep *ep = ctx;
1208 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1210 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1211 rpl->status, status2errno(rpl->status));
1212 ep->com.rpl_err = status2errno(rpl->status);
1213 ep->com.rpl_done = 1;
1214 wake_up(&ep->com.waitq);
1216 return CPL_RET_BUF_DONE;
1219 static int listen_stop(struct iwch_listen_ep *ep)
1221 struct sk_buff *skb;
1222 struct cpl_close_listserv_req *req;
1224 PDBG("%s ep %p\n", __func__, ep);
1225 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1227 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1230 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1231 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1233 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1235 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1238 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1241 struct iwch_listen_ep *ep = ctx;
1242 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1244 PDBG("%s ep %p\n", __func__, ep);
1245 ep->com.rpl_err = status2errno(rpl->status);
1246 ep->com.rpl_done = 1;
1247 wake_up(&ep->com.waitq);
1248 return CPL_RET_BUF_DONE;
1251 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1253 struct cpl_pass_accept_rpl *rpl;
1254 unsigned int mtu_idx;
1255 u32 opt0h, opt0l, opt2;
1258 PDBG("%s ep %p\n", __func__, ep);
1259 BUG_ON(skb_cloned(skb));
1260 skb_trim(skb, sizeof(*rpl));
1262 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1263 wscale = compute_wscale(rcv_win);
1264 opt0h = V_NAGLE(0) |
1268 V_WND_SCALE(wscale) |
1269 V_MSS_IDX(mtu_idx) |
1270 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1271 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1272 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1273 V_CONG_CONTROL_FLAVOR(cong_flavor);
1276 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1277 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1278 rpl->peer_ip = peer_ip;
1279 rpl->opt0h = htonl(opt0h);
1280 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1281 rpl->opt2 = htonl(opt2);
1282 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1283 skb->priority = CPL_PRIORITY_SETUP;
1284 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1289 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1290 struct sk_buff *skb)
1292 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1294 BUG_ON(skb_cloned(skb));
1295 skb_trim(skb, sizeof(struct cpl_tid_release));
1298 if (tdev->type != T3A)
1299 release_tid(tdev, hwtid, skb);
1301 struct cpl_pass_accept_rpl *rpl;
1304 skb->priority = CPL_PRIORITY_SETUP;
1305 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1306 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1308 rpl->peer_ip = peer_ip;
1309 rpl->opt0h = htonl(F_TCAM_BYPASS);
1310 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1312 rpl->rsvd = rpl->opt2;
1313 iwch_cxgb3_ofld_send(tdev, skb);
1317 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1319 struct iwch_ep *child_ep, *parent_ep = ctx;
1320 struct cpl_pass_accept_req *req = cplhdr(skb);
1321 unsigned int hwtid = GET_TID(req);
1322 struct dst_entry *dst;
1323 struct l2t_entry *l2t;
1327 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1329 if (state_read(&parent_ep->com) != LISTEN) {
1330 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1336 * Find the netdev for this connection request.
1338 tim.mac_addr = req->dst_mac;
1339 tim.vlan_tag = ntohs(req->vlan_tag);
1340 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1341 printk(KERN_ERR "%s bad dst mac %pM\n",
1342 __func__, req->dst_mac);
1346 /* Find output route */
1347 rt = find_route(tdev,
1351 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1353 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1358 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1360 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1365 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1367 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1369 l2t_release(L2DATA(tdev), l2t);
1373 state_set(&child_ep->com, CONNECTING);
1374 child_ep->com.tdev = tdev;
1375 child_ep->com.cm_id = NULL;
1376 child_ep->com.local_addr.sin_family = PF_INET;
1377 child_ep->com.local_addr.sin_port = req->local_port;
1378 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1379 child_ep->com.remote_addr.sin_family = PF_INET;
1380 child_ep->com.remote_addr.sin_port = req->peer_port;
1381 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1382 get_ep(&parent_ep->com);
1383 child_ep->parent_ep = parent_ep;
1384 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1385 child_ep->l2t = l2t;
1386 child_ep->dst = dst;
1387 child_ep->hwtid = hwtid;
1388 init_timer(&child_ep->timer);
1389 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1390 accept_cr(child_ep, req->peer_ip, skb);
1393 reject_cr(tdev, hwtid, req->peer_ip, skb);
1395 return CPL_RET_BUF_DONE;
1398 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1400 struct iwch_ep *ep = ctx;
1401 struct cpl_pass_establish *req = cplhdr(skb);
1403 PDBG("%s ep %p\n", __func__, ep);
1404 ep->snd_seq = ntohl(req->snd_isn);
1405 ep->rcv_seq = ntohl(req->rcv_isn);
1407 set_emss(ep, ntohs(req->tcp_opt));
1409 dst_confirm(ep->dst);
1410 state_set(&ep->com, MPA_REQ_WAIT);
1413 return CPL_RET_BUF_DONE;
1416 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1418 struct iwch_ep *ep = ctx;
1419 struct iwch_qp_attributes attrs;
1420 unsigned long flags;
1424 PDBG("%s ep %p\n", __func__, ep);
1425 dst_confirm(ep->dst);
1427 spin_lock_irqsave(&ep->com.lock, flags);
1428 switch (ep->com.state) {
1430 __state_set(&ep->com, CLOSING);
1433 __state_set(&ep->com, CLOSING);
1434 connect_reply_upcall(ep, -ECONNRESET);
1439 * We're gonna mark this puppy DEAD, but keep
1440 * the reference on it until the ULP accepts or
1441 * rejects the CR. Also wake up anyone waiting
1442 * in rdma connection migration (see iwch_accept_cr()).
1444 __state_set(&ep->com, CLOSING);
1445 ep->com.rpl_done = 1;
1446 ep->com.rpl_err = -ECONNRESET;
1447 PDBG("waking up ep %p\n", ep);
1448 wake_up(&ep->com.waitq);
1451 __state_set(&ep->com, CLOSING);
1452 ep->com.rpl_done = 1;
1453 ep->com.rpl_err = -ECONNRESET;
1454 PDBG("waking up ep %p\n", ep);
1455 wake_up(&ep->com.waitq);
1459 __state_set(&ep->com, CLOSING);
1460 attrs.next_state = IWCH_QP_STATE_CLOSING;
1461 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1462 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1463 peer_close_upcall(ep);
1469 __state_set(&ep->com, MORIBUND);
1474 if (ep->com.cm_id && ep->com.qp) {
1475 attrs.next_state = IWCH_QP_STATE_IDLE;
1476 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1477 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1479 close_complete_upcall(ep);
1480 __state_set(&ep->com, DEAD);
1490 spin_unlock_irqrestore(&ep->com.lock, flags);
1492 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1494 release_ep_resources(ep);
1495 return CPL_RET_BUF_DONE;
1499 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1501 static int is_neg_adv_abort(unsigned int status)
1503 return status == CPL_ERR_RTX_NEG_ADVICE ||
1504 status == CPL_ERR_PERSIST_NEG_ADVICE;
1507 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1509 struct cpl_abort_req_rss *req = cplhdr(skb);
1510 struct iwch_ep *ep = ctx;
1511 struct cpl_abort_rpl *rpl;
1512 struct sk_buff *rpl_skb;
1513 struct iwch_qp_attributes attrs;
1516 unsigned long flags;
1518 if (is_neg_adv_abort(req->status)) {
1519 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1521 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1522 return CPL_RET_BUF_DONE;
1526 * We get 2 peer aborts from the HW. The first one must
1527 * be ignored except for scribbling that we need one more.
1529 if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
1530 return CPL_RET_BUF_DONE;
1533 spin_lock_irqsave(&ep->com.lock, flags);
1534 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1535 switch (ep->com.state) {
1543 connect_reply_upcall(ep, -ECONNRESET);
1546 ep->com.rpl_done = 1;
1547 ep->com.rpl_err = -ECONNRESET;
1548 PDBG("waking up ep %p\n", ep);
1549 wake_up(&ep->com.waitq);
1554 * We're gonna mark this puppy DEAD, but keep
1555 * the reference on it until the ULP accepts or
1556 * rejects the CR. Also wake up anyone waiting
1557 * in rdma connection migration (see iwch_accept_cr()).
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);
1569 if (ep->com.cm_id && ep->com.qp) {
1570 attrs.next_state = IWCH_QP_STATE_ERROR;
1571 ret = iwch_modify_qp(ep->com.qp->rhp,
1572 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1576 "%s - qp <- error failed!\n",
1579 peer_abort_upcall(ep);
1584 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1585 spin_unlock_irqrestore(&ep->com.lock, flags);
1586 return CPL_RET_BUF_DONE;
1591 dst_confirm(ep->dst);
1592 if (ep->com.state != ABORTING) {
1593 __state_set(&ep->com, DEAD);
1596 spin_unlock_irqrestore(&ep->com.lock, flags);
1598 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1600 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1605 rpl_skb->priority = CPL_PRIORITY_DATA;
1606 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1607 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1608 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1609 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1610 rpl->cmd = CPL_ABORT_NO_RST;
1611 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1614 release_ep_resources(ep);
1615 return CPL_RET_BUF_DONE;
1618 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1620 struct iwch_ep *ep = ctx;
1621 struct iwch_qp_attributes attrs;
1622 unsigned long flags;
1625 PDBG("%s ep %p\n", __func__, ep);
1628 /* The cm_id may be null if we failed to connect */
1629 spin_lock_irqsave(&ep->com.lock, flags);
1630 switch (ep->com.state) {
1632 __state_set(&ep->com, MORIBUND);
1636 if ((ep->com.cm_id) && (ep->com.qp)) {
1637 attrs.next_state = IWCH_QP_STATE_IDLE;
1638 iwch_modify_qp(ep->com.qp->rhp,
1640 IWCH_QP_ATTR_NEXT_STATE,
1643 close_complete_upcall(ep);
1644 __state_set(&ep->com, DEAD);
1654 spin_unlock_irqrestore(&ep->com.lock, flags);
1656 release_ep_resources(ep);
1657 return CPL_RET_BUF_DONE;
1661 * T3A does 3 things when a TERM is received:
1662 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1663 * 2) generate an async event on the QP with the TERMINATE opcode
1664 * 3) post a TERMINATE opcde cqe into the associated CQ.
1666 * For (1), we save the message in the qp for later consumer consumption.
1667 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1668 * For (3), we toss the CQE in cxio_poll_cq().
1670 * terminate() handles case (1)...
1672 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1674 struct iwch_ep *ep = ctx;
1676 if (state_read(&ep->com) != FPDU_MODE)
1677 return CPL_RET_BUF_DONE;
1679 PDBG("%s ep %p\n", __func__, ep);
1680 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1681 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1682 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1684 ep->com.qp->attr.terminate_msg_len = skb->len;
1685 ep->com.qp->attr.is_terminate_local = 0;
1686 return CPL_RET_BUF_DONE;
1689 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1691 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1692 struct iwch_ep *ep = ctx;
1694 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
1697 struct iwch_qp_attributes attrs;
1699 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1700 __func__, ep->hwtid);
1702 attrs.next_state = IWCH_QP_STATE_ERROR;
1703 iwch_modify_qp(ep->com.qp->rhp,
1704 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1706 abort_connection(ep, NULL, GFP_KERNEL);
1708 return CPL_RET_BUF_DONE;
1711 static void ep_timeout(unsigned long arg)
1713 struct iwch_ep *ep = (struct iwch_ep *)arg;
1714 struct iwch_qp_attributes attrs;
1715 unsigned long flags;
1718 spin_lock_irqsave(&ep->com.lock, flags);
1719 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
1721 switch (ep->com.state) {
1723 __state_set(&ep->com, ABORTING);
1724 connect_reply_upcall(ep, -ETIMEDOUT);
1727 __state_set(&ep->com, ABORTING);
1731 if (ep->com.cm_id && ep->com.qp) {
1732 attrs.next_state = IWCH_QP_STATE_ERROR;
1733 iwch_modify_qp(ep->com.qp->rhp,
1734 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1737 __state_set(&ep->com, ABORTING);
1740 printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1741 __func__, ep, ep->com.state);
1745 spin_unlock_irqrestore(&ep->com.lock, flags);
1747 abort_connection(ep, NULL, GFP_ATOMIC);
1751 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1754 struct iwch_ep *ep = to_ep(cm_id);
1755 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1757 if (state_read(&ep->com) == DEAD) {
1761 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1763 abort_connection(ep, NULL, GFP_KERNEL);
1765 err = send_mpa_reject(ep, pdata, pdata_len);
1766 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1772 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1775 struct iwch_qp_attributes attrs;
1776 enum iwch_qp_attr_mask mask;
1777 struct iwch_ep *ep = to_ep(cm_id);
1778 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1779 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1781 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1782 if (state_read(&ep->com) == DEAD) {
1787 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1790 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1791 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1792 abort_connection(ep, NULL, GFP_KERNEL);
1797 cm_id->add_ref(cm_id);
1798 ep->com.cm_id = cm_id;
1801 ep->ird = conn_param->ird;
1802 ep->ord = conn_param->ord;
1804 if (peer2peer && ep->ird == 0)
1807 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1809 /* bind QP to EP and move to RTS */
1810 attrs.mpa_attr = ep->mpa_attr;
1811 attrs.max_ird = ep->ird;
1812 attrs.max_ord = ep->ord;
1813 attrs.llp_stream_handle = ep;
1814 attrs.next_state = IWCH_QP_STATE_RTS;
1816 /* bind QP and TID with INIT_WR */
1817 mask = IWCH_QP_ATTR_NEXT_STATE |
1818 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1819 IWCH_QP_ATTR_MPA_ATTR |
1820 IWCH_QP_ATTR_MAX_IRD |
1821 IWCH_QP_ATTR_MAX_ORD;
1823 err = iwch_modify_qp(ep->com.qp->rhp,
1824 ep->com.qp, mask, &attrs, 1);
1828 /* if needed, wait for wr_ack */
1829 if (iwch_rqes_posted(qp)) {
1830 wait_event(ep->com.waitq, ep->com.rpl_done);
1831 err = ep->com.rpl_err;
1836 err = send_mpa_reply(ep, conn_param->private_data,
1837 conn_param->private_data_len);
1842 state_set(&ep->com, FPDU_MODE);
1843 established_upcall(ep);
1847 ep->com.cm_id = NULL;
1849 cm_id->rem_ref(cm_id);
1855 static int is_loopback_dst(struct iw_cm_id *cm_id)
1857 struct net_device *dev;
1859 dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1866 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1869 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1873 if (is_loopback_dst(cm_id)) {
1878 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1880 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1884 init_timer(&ep->timer);
1885 ep->plen = conn_param->private_data_len;
1887 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1888 conn_param->private_data, ep->plen);
1889 ep->ird = conn_param->ird;
1890 ep->ord = conn_param->ord;
1892 if (peer2peer && ep->ord == 0)
1895 ep->com.tdev = h->rdev.t3cdev_p;
1897 cm_id->add_ref(cm_id);
1898 ep->com.cm_id = cm_id;
1899 ep->com.qp = get_qhp(h, conn_param->qpn);
1900 BUG_ON(!ep->com.qp);
1901 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1905 * Allocate an active TID to initiate a TCP connection.
1907 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1908 if (ep->atid == -1) {
1909 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1915 rt = find_route(h->rdev.t3cdev_p,
1916 cm_id->local_addr.sin_addr.s_addr,
1917 cm_id->remote_addr.sin_addr.s_addr,
1918 cm_id->local_addr.sin_port,
1919 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1921 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1922 err = -EHOSTUNREACH;
1927 /* get a l2t entry */
1928 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1929 ep->dst->neighbour->dev);
1931 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1936 state_set(&ep->com, CONNECTING);
1937 ep->tos = IPTOS_LOWDELAY;
1938 ep->com.local_addr = cm_id->local_addr;
1939 ep->com.remote_addr = cm_id->remote_addr;
1941 /* send connect request to rnic */
1942 err = send_connect(ep);
1946 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1948 dst_release(ep->dst);
1950 cxgb3_free_atid(ep->com.tdev, ep->atid);
1952 cm_id->rem_ref(cm_id);
1958 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1961 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1962 struct iwch_listen_ep *ep;
1967 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1969 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1973 PDBG("%s ep %p\n", __func__, ep);
1974 ep->com.tdev = h->rdev.t3cdev_p;
1975 cm_id->add_ref(cm_id);
1976 ep->com.cm_id = cm_id;
1977 ep->backlog = backlog;
1978 ep->com.local_addr = cm_id->local_addr;
1981 * Allocate a server TID.
1983 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1984 if (ep->stid == -1) {
1985 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1990 state_set(&ep->com, LISTEN);
1991 err = listen_start(ep);
1995 /* wait for pass_open_rpl */
1996 wait_event(ep->com.waitq, ep->com.rpl_done);
1997 err = ep->com.rpl_err;
1999 cm_id->provider_data = ep;
2003 cxgb3_free_stid(ep->com.tdev, ep->stid);
2005 cm_id->rem_ref(cm_id);
2012 int iwch_destroy_listen(struct iw_cm_id *cm_id)
2015 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2017 PDBG("%s ep %p\n", __func__, ep);
2020 state_set(&ep->com, DEAD);
2021 ep->com.rpl_done = 0;
2022 ep->com.rpl_err = 0;
2023 err = listen_stop(ep);
2026 wait_event(ep->com.waitq, ep->com.rpl_done);
2027 cxgb3_free_stid(ep->com.tdev, ep->stid);
2029 err = ep->com.rpl_err;
2030 cm_id->rem_ref(cm_id);
2035 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2038 unsigned long flags;
2041 struct t3cdev *tdev;
2042 struct cxio_rdev *rdev;
2044 spin_lock_irqsave(&ep->com.lock, flags);
2046 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2047 states[ep->com.state], abrupt);
2049 tdev = (struct t3cdev *)ep->com.tdev;
2050 rdev = (struct cxio_rdev *)tdev->ulp;
2051 if (cxio_fatal_error(rdev)) {
2053 close_complete_upcall(ep);
2054 ep->com.state = DEAD;
2056 switch (ep->com.state) {
2064 ep->com.state = ABORTING;
2066 ep->com.state = CLOSING;
2069 set_bit(CLOSE_SENT, &ep->com.flags);
2072 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2076 ep->com.state = ABORTING;
2078 ep->com.state = MORIBUND;
2084 PDBG("%s ignoring disconnect ep %p state %u\n",
2085 __func__, ep, ep->com.state);
2092 spin_unlock_irqrestore(&ep->com.lock, flags);
2095 ret = send_abort(ep, NULL, gfp);
2097 ret = send_halfclose(ep, gfp);
2102 release_ep_resources(ep);
2106 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2107 struct l2t_entry *l2t)
2109 struct iwch_ep *ep = ctx;
2114 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2117 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2125 * All the CM events are handled on a work queue to have a safe context.
2126 * These are the real handlers that are called from the work queue.
2128 static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2129 [CPL_ACT_ESTABLISH] = act_establish,
2130 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2131 [CPL_RX_DATA] = rx_data,
2132 [CPL_TX_DMA_ACK] = tx_ack,
2133 [CPL_ABORT_RPL_RSS] = abort_rpl,
2134 [CPL_ABORT_RPL] = abort_rpl,
2135 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2136 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2137 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2138 [CPL_PASS_ESTABLISH] = pass_establish,
2139 [CPL_PEER_CLOSE] = peer_close,
2140 [CPL_ABORT_REQ_RSS] = peer_abort,
2141 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2142 [CPL_RDMA_TERMINATE] = terminate,
2143 [CPL_RDMA_EC_STATUS] = ec_status,
2146 static void process_work(struct work_struct *work)
2148 struct sk_buff *skb = NULL;
2150 struct t3cdev *tdev;
2153 while ((skb = skb_dequeue(&rxq))) {
2154 ep = *((void **) (skb->cb));
2155 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2156 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2157 if (ret & CPL_RET_BUF_DONE)
2161 * ep was referenced in sched(), and is freed here.
2163 put_ep((struct iwch_ep_common *)ep);
2167 static DECLARE_WORK(skb_work, process_work);
2169 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2171 struct iwch_ep_common *epc = ctx;
2176 * Save ctx and tdev in the skb->cb area.
2178 *((void **) skb->cb) = ctx;
2179 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2182 * Queue the skb and schedule the worker thread.
2184 skb_queue_tail(&rxq, skb);
2185 queue_work(workq, &skb_work);
2189 static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2191 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2193 if (rpl->status != CPL_ERR_NONE) {
2194 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2195 "for tid %u\n", rpl->status, GET_TID(rpl));
2197 return CPL_RET_BUF_DONE;
2201 * All upcalls from the T3 Core go to sched() to schedule the
2202 * processing on a work queue.
2204 cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2205 [CPL_ACT_ESTABLISH] = sched,
2206 [CPL_ACT_OPEN_RPL] = sched,
2207 [CPL_RX_DATA] = sched,
2208 [CPL_TX_DMA_ACK] = sched,
2209 [CPL_ABORT_RPL_RSS] = sched,
2210 [CPL_ABORT_RPL] = sched,
2211 [CPL_PASS_OPEN_RPL] = sched,
2212 [CPL_CLOSE_LISTSRV_RPL] = sched,
2213 [CPL_PASS_ACCEPT_REQ] = sched,
2214 [CPL_PASS_ESTABLISH] = sched,
2215 [CPL_PEER_CLOSE] = sched,
2216 [CPL_CLOSE_CON_RPL] = sched,
2217 [CPL_ABORT_REQ_RSS] = sched,
2218 [CPL_RDMA_TERMINATE] = sched,
2219 [CPL_RDMA_EC_STATUS] = sched,
2220 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2223 int __init iwch_cm_init(void)
2225 skb_queue_head_init(&rxq);
2227 workq = create_singlethread_workqueue("iw_cxgb3");
2234 void __exit iwch_cm_term(void)
2236 flush_workqueue(workq);
2237 destroy_workqueue(workq);