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/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
39 #include <net/neighbour.h>
40 #include <net/netevent.h>
41 #include <net/route.h>
44 #include "cxgb3_offload.h"
46 #include "iwch_provider.h"
49 static char *states[] = {
65 static int ep_timeout_secs = 10;
66 module_param(ep_timeout_secs, int, 0444);
67 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
68 "in seconds (default=10)");
70 static int mpa_rev = 1;
71 module_param(mpa_rev, int, 0444);
72 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
73 "1 is spec compliant. (default=1)");
75 static int markers_enabled = 0;
76 module_param(markers_enabled, int, 0444);
77 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
79 static int crc_enabled = 1;
80 module_param(crc_enabled, int, 0444);
81 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
83 static int rcv_win = 256 * 1024;
84 module_param(rcv_win, int, 0444);
85 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
87 static int snd_win = 32 * 1024;
88 module_param(snd_win, int, 0444);
89 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
91 static unsigned int nocong = 0;
92 module_param(nocong, uint, 0444);
93 MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
95 static unsigned int cong_flavor = 1;
96 module_param(cong_flavor, uint, 0444);
97 MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
99 static void process_work(struct work_struct *work);
100 static struct workqueue_struct *workq;
101 static DECLARE_WORK(skb_work, process_work);
103 static struct sk_buff_head rxq;
104 static cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS];
106 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
107 static void ep_timeout(unsigned long arg);
108 static void connect_reply_upcall(struct iwch_ep *ep, int status);
110 static void start_ep_timer(struct iwch_ep *ep)
112 PDBG("%s ep %p\n", __FUNCTION__, ep);
113 if (timer_pending(&ep->timer)) {
114 PDBG("%s stopped / restarted timer ep %p\n", __FUNCTION__, ep);
115 del_timer_sync(&ep->timer);
118 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
119 ep->timer.data = (unsigned long)ep;
120 ep->timer.function = ep_timeout;
121 add_timer(&ep->timer);
124 static void stop_ep_timer(struct iwch_ep *ep)
126 PDBG("%s ep %p\n", __FUNCTION__, ep);
127 del_timer_sync(&ep->timer);
131 static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
133 struct cpl_tid_release *req;
135 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
138 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
139 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
140 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
141 skb->priority = CPL_PRIORITY_SETUP;
142 tdev->send(tdev, skb);
146 int iwch_quiesce_tid(struct iwch_ep *ep)
148 struct cpl_set_tcb_field *req;
149 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
153 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
154 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
155 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
156 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
159 req->word = htons(W_TCB_RX_QUIESCE);
160 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
161 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
163 skb->priority = CPL_PRIORITY_DATA;
164 ep->com.tdev->send(ep->com.tdev, skb);
168 int iwch_resume_tid(struct iwch_ep *ep)
170 struct cpl_set_tcb_field *req;
171 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
175 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
176 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
177 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
178 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
181 req->word = htons(W_TCB_RX_QUIESCE);
182 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
185 skb->priority = CPL_PRIORITY_DATA;
186 ep->com.tdev->send(ep->com.tdev, skb);
190 static void set_emss(struct iwch_ep *ep, u16 opt)
192 PDBG("%s ep %p opt %u\n", __FUNCTION__, ep, opt);
193 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
194 if (G_TCPOPT_TSTAMP(opt))
198 PDBG("emss=%d\n", ep->emss);
201 static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
204 enum iwch_ep_state state;
206 spin_lock_irqsave(&epc->lock, flags);
208 spin_unlock_irqrestore(&epc->lock, flags);
212 static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
217 static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
221 spin_lock_irqsave(&epc->lock, flags);
222 PDBG("%s - %s -> %s\n", __FUNCTION__, states[epc->state], states[new]);
223 __state_set(epc, new);
224 spin_unlock_irqrestore(&epc->lock, flags);
228 static void *alloc_ep(int size, gfp_t gfp)
230 struct iwch_ep_common *epc;
232 epc = kmalloc(size, gfp);
234 memset(epc, 0, size);
235 kref_init(&epc->kref);
236 spin_lock_init(&epc->lock);
237 init_waitqueue_head(&epc->waitq);
239 PDBG("%s alloc ep %p\n", __FUNCTION__, epc);
243 void __free_ep(struct kref *kref)
245 struct iwch_ep_common *epc;
246 epc = container_of(kref, struct iwch_ep_common, kref);
247 PDBG("%s ep %p state %s\n", __FUNCTION__, epc, states[state_read(epc)]);
251 static void release_ep_resources(struct iwch_ep *ep)
253 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
254 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
255 dst_release(ep->dst);
256 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
257 if (ep->com.tdev->type == T3B)
258 release_tid(ep->com.tdev, ep->hwtid, NULL);
262 static void process_work(struct work_struct *work)
264 struct sk_buff *skb = NULL;
269 while ((skb = skb_dequeue(&rxq))) {
270 ep = *((void **) (skb->cb));
271 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
272 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
273 if (ret & CPL_RET_BUF_DONE)
277 * ep was referenced in sched(), and is freed here.
279 put_ep((struct iwch_ep_common *)ep);
283 static int status2errno(int status)
288 case CPL_ERR_CONN_RESET:
290 case CPL_ERR_ARP_MISS:
291 return -EHOSTUNREACH;
292 case CPL_ERR_CONN_TIMEDOUT:
294 case CPL_ERR_TCAM_FULL:
296 case CPL_ERR_CONN_EXIST:
304 * Try and reuse skbs already allocated...
306 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
308 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
312 skb = alloc_skb(len, gfp);
317 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
318 __be32 peer_ip, __be16 local_port,
319 __be16 peer_port, u8 tos)
330 .proto = IPPROTO_TCP,
338 if (ip_route_output_flow(&rt, &fl, NULL, 0))
343 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
347 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
352 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
354 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
359 * Handle an ARP failure for an active open.
361 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
363 printk(KERN_ERR MOD "ARP failure duing connect\n");
368 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
371 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
373 struct cpl_abort_req *req = cplhdr(skb);
375 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
376 req->cmd = CPL_ABORT_NO_RST;
377 cxgb3_ofld_send(dev, skb);
380 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
382 struct cpl_close_con_req *req;
385 PDBG("%s ep %p\n", __FUNCTION__, ep);
386 skb = get_skb(NULL, sizeof(*req), gfp);
388 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
391 skb->priority = CPL_PRIORITY_DATA;
392 set_arp_failure_handler(skb, arp_failure_discard);
393 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
394 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
395 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
396 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
397 l2t_send(ep->com.tdev, skb, ep->l2t);
401 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
403 struct cpl_abort_req *req;
405 PDBG("%s ep %p\n", __FUNCTION__, ep);
406 skb = get_skb(skb, sizeof(*req), gfp);
408 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
412 skb->priority = CPL_PRIORITY_DATA;
413 set_arp_failure_handler(skb, abort_arp_failure);
414 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
415 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
416 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
417 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
418 req->cmd = CPL_ABORT_SEND_RST;
419 l2t_send(ep->com.tdev, skb, ep->l2t);
423 static int send_connect(struct iwch_ep *ep)
425 struct cpl_act_open_req *req;
427 u32 opt0h, opt0l, opt2;
428 unsigned int mtu_idx;
431 PDBG("%s ep %p\n", __FUNCTION__, ep);
433 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
435 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
439 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
440 wscale = compute_wscale(rcv_win);
445 V_WND_SCALE(wscale) |
447 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
448 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
449 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
450 skb->priority = CPL_PRIORITY_SETUP;
451 set_arp_failure_handler(skb, act_open_req_arp_failure);
453 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
454 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
455 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
456 req->local_port = ep->com.local_addr.sin_port;
457 req->peer_port = ep->com.remote_addr.sin_port;
458 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
459 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
460 req->opt0h = htonl(opt0h);
461 req->opt0l = htonl(opt0l);
463 req->opt2 = htonl(opt2);
464 l2t_send(ep->com.tdev, skb, ep->l2t);
468 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
471 struct tx_data_wr *req;
472 struct mpa_message *mpa;
475 PDBG("%s ep %p pd_len %d\n", __FUNCTION__, ep, ep->plen);
477 BUG_ON(skb_cloned(skb));
479 mpalen = sizeof(*mpa) + ep->plen;
480 if (skb->data + mpalen + sizeof(*req) > skb->end) {
482 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
484 connect_reply_upcall(ep, -ENOMEM);
489 skb_reserve(skb, sizeof(*req));
490 skb_put(skb, mpalen);
491 skb->priority = CPL_PRIORITY_DATA;
492 mpa = (struct mpa_message *) skb->data;
493 memset(mpa, 0, sizeof(*mpa));
494 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
495 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
496 (markers_enabled ? MPA_MARKERS : 0);
497 mpa->private_data_size = htons(ep->plen);
498 mpa->revision = mpa_rev;
501 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
504 * Reference the mpa skb. This ensures the data area
505 * will remain in memory until the hw acks the tx.
506 * Function tx_ack() will deref it.
509 set_arp_failure_handler(skb, arp_failure_discard);
510 skb->h.raw = skb->data;
512 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
513 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
514 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
515 req->len = htonl(len);
516 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
517 V_TX_SNDBUF(snd_win>>15));
518 req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
519 req->sndseq = htonl(ep->snd_seq);
522 l2t_send(ep->com.tdev, skb, ep->l2t);
524 state_set(&ep->com, MPA_REQ_SENT);
528 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
531 struct tx_data_wr *req;
532 struct mpa_message *mpa;
535 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
537 mpalen = sizeof(*mpa) + plen;
539 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
541 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
544 skb_reserve(skb, sizeof(*req));
545 mpa = (struct mpa_message *) skb_put(skb, mpalen);
546 memset(mpa, 0, sizeof(*mpa));
547 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
548 mpa->flags = MPA_REJECT;
549 mpa->revision = mpa_rev;
550 mpa->private_data_size = htons(plen);
552 memcpy(mpa->private_data, pdata, plen);
555 * Reference the mpa skb again. This ensures the data area
556 * will remain in memory until the hw acks the tx.
557 * Function tx_ack() will deref it.
560 skb->priority = CPL_PRIORITY_DATA;
561 set_arp_failure_handler(skb, arp_failure_discard);
562 skb->h.raw = skb->data;
563 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
564 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
565 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
566 req->len = htonl(mpalen);
567 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
568 V_TX_SNDBUF(snd_win>>15));
569 req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
570 req->sndseq = htonl(ep->snd_seq);
573 l2t_send(ep->com.tdev, skb, ep->l2t);
577 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
580 struct tx_data_wr *req;
581 struct mpa_message *mpa;
585 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
587 mpalen = sizeof(*mpa) + plen;
589 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
591 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
594 skb->priority = CPL_PRIORITY_DATA;
595 skb_reserve(skb, sizeof(*req));
596 mpa = (struct mpa_message *) skb_put(skb, mpalen);
597 memset(mpa, 0, sizeof(*mpa));
598 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
599 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
600 (markers_enabled ? MPA_MARKERS : 0);
601 mpa->revision = mpa_rev;
602 mpa->private_data_size = htons(plen);
604 memcpy(mpa->private_data, pdata, plen);
607 * Reference the mpa skb. This ensures the data area
608 * will remain in memory until the hw acks the tx.
609 * Function tx_ack() will deref it.
612 set_arp_failure_handler(skb, arp_failure_discard);
613 skb->h.raw = skb->data;
615 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
616 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
617 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
618 req->len = htonl(len);
619 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
620 V_TX_SNDBUF(snd_win>>15));
621 req->flags = htonl(F_TX_MORE | F_TX_IMM_ACK | F_TX_INIT);
622 req->sndseq = htonl(ep->snd_seq);
624 state_set(&ep->com, MPA_REP_SENT);
625 l2t_send(ep->com.tdev, skb, ep->l2t);
629 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
631 struct iwch_ep *ep = ctx;
632 struct cpl_act_establish *req = cplhdr(skb);
633 unsigned int tid = GET_TID(req);
635 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, tid);
637 dst_confirm(ep->dst);
639 /* setup the hwtid for this connection */
641 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
643 ep->snd_seq = ntohl(req->snd_isn);
645 set_emss(ep, ntohs(req->tcp_opt));
647 /* dealloc the atid */
648 cxgb3_free_atid(ep->com.tdev, ep->atid);
650 /* start MPA negotiation */
651 send_mpa_req(ep, skb);
656 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
658 PDBG("%s ep %p\n", __FILE__, ep);
659 state_set(&ep->com, ABORTING);
660 send_abort(ep, skb, gfp);
663 static void close_complete_upcall(struct iwch_ep *ep)
665 struct iw_cm_event event;
667 PDBG("%s ep %p\n", __FUNCTION__, ep);
668 memset(&event, 0, sizeof(event));
669 event.event = IW_CM_EVENT_CLOSE;
671 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
672 ep, ep->com.cm_id, ep->hwtid);
673 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
674 ep->com.cm_id->rem_ref(ep->com.cm_id);
675 ep->com.cm_id = NULL;
680 static void peer_close_upcall(struct iwch_ep *ep)
682 struct iw_cm_event event;
684 PDBG("%s ep %p\n", __FUNCTION__, ep);
685 memset(&event, 0, sizeof(event));
686 event.event = IW_CM_EVENT_DISCONNECT;
688 PDBG("peer close 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);
694 static void peer_abort_upcall(struct iwch_ep *ep)
696 struct iw_cm_event event;
698 PDBG("%s ep %p\n", __FUNCTION__, ep);
699 memset(&event, 0, sizeof(event));
700 event.event = IW_CM_EVENT_CLOSE;
701 event.status = -ECONNRESET;
703 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
704 ep->com.cm_id, ep->hwtid);
705 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
706 ep->com.cm_id->rem_ref(ep->com.cm_id);
707 ep->com.cm_id = NULL;
712 static void connect_reply_upcall(struct iwch_ep *ep, int status)
714 struct iw_cm_event event;
716 PDBG("%s ep %p status %d\n", __FUNCTION__, ep, status);
717 memset(&event, 0, sizeof(event));
718 event.event = IW_CM_EVENT_CONNECT_REPLY;
719 event.status = status;
720 event.local_addr = ep->com.local_addr;
721 event.remote_addr = ep->com.remote_addr;
723 if ((status == 0) || (status == -ECONNREFUSED)) {
724 event.private_data_len = ep->plen;
725 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
728 PDBG("%s ep %p tid %d status %d\n", __FUNCTION__, ep,
730 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
733 ep->com.cm_id->rem_ref(ep->com.cm_id);
734 ep->com.cm_id = NULL;
739 static void connect_request_upcall(struct iwch_ep *ep)
741 struct iw_cm_event event;
743 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
744 memset(&event, 0, sizeof(event));
745 event.event = IW_CM_EVENT_CONNECT_REQUEST;
746 event.local_addr = ep->com.local_addr;
747 event.remote_addr = ep->com.remote_addr;
748 event.private_data_len = ep->plen;
749 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
750 event.provider_data = ep;
751 if (state_read(&ep->parent_ep->com) != DEAD)
752 ep->parent_ep->com.cm_id->event_handler(
753 ep->parent_ep->com.cm_id,
755 put_ep(&ep->parent_ep->com);
756 ep->parent_ep = NULL;
759 static void established_upcall(struct iwch_ep *ep)
761 struct iw_cm_event event;
763 PDBG("%s ep %p\n", __FUNCTION__, ep);
764 memset(&event, 0, sizeof(event));
765 event.event = IW_CM_EVENT_ESTABLISHED;
767 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
768 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
772 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
774 struct cpl_rx_data_ack *req;
777 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
778 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
780 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
784 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
785 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
786 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
787 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
788 skb->priority = CPL_PRIORITY_ACK;
789 ep->com.tdev->send(ep->com.tdev, skb);
793 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
795 struct mpa_message *mpa;
797 struct iwch_qp_attributes attrs;
798 enum iwch_qp_attr_mask mask;
801 PDBG("%s ep %p\n", __FUNCTION__, ep);
804 * Stop mpa timer. If it expired, then the state has
805 * changed and we bail since ep_timeout already aborted
809 if (state_read(&ep->com) != MPA_REQ_SENT)
813 * If we get more than the supported amount of private data
814 * then we must fail this connection.
816 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
822 * copy the new data into our accumulation buffer.
824 memcpy(&(ep->mpa_pkt[ep->mpa_pkt_len]), skb->data, skb->len);
825 ep->mpa_pkt_len += skb->len;
828 * if we don't even have the mpa message, then bail.
830 if (ep->mpa_pkt_len < sizeof(*mpa))
832 mpa = (struct mpa_message *) ep->mpa_pkt;
834 /* Validate MPA header. */
835 if (mpa->revision != mpa_rev) {
839 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
844 plen = ntohs(mpa->private_data_size);
847 * Fail if there's too much private data.
849 if (plen > MPA_MAX_PRIVATE_DATA) {
855 * If plen does not account for pkt size
857 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
862 ep->plen = (u8) plen;
865 * If we don't have all the pdata yet, then bail.
866 * We'll continue process when more data arrives.
868 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
871 if (mpa->flags & MPA_REJECT) {
877 * If we get here we have accumulated the entire mpa
878 * start reply message including private data. And
879 * the MPA header is valid.
881 state_set(&ep->com, FPDU_MODE);
882 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
883 ep->mpa_attr.recv_marker_enabled = markers_enabled;
884 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
885 ep->mpa_attr.version = mpa_rev;
886 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
887 "xmit_marker_enabled=%d, version=%d\n", __FUNCTION__,
888 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
889 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
891 attrs.mpa_attr = ep->mpa_attr;
892 attrs.max_ird = ep->ird;
893 attrs.max_ord = ep->ord;
894 attrs.llp_stream_handle = ep;
895 attrs.next_state = IWCH_QP_STATE_RTS;
897 mask = IWCH_QP_ATTR_NEXT_STATE |
898 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
899 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
901 /* bind QP and TID with INIT_WR */
902 err = iwch_modify_qp(ep->com.qp->rhp,
903 ep->com.qp, mask, &attrs, 1);
907 abort_connection(ep, skb, GFP_KERNEL);
909 connect_reply_upcall(ep, err);
913 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
915 struct mpa_message *mpa;
918 PDBG("%s ep %p\n", __FUNCTION__, ep);
921 * Stop mpa timer. If it expired, then the state has
922 * changed and we bail since ep_timeout already aborted
926 if (state_read(&ep->com) != MPA_REQ_WAIT)
930 * If we get more than the supported amount of private data
931 * then we must fail this connection.
933 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
934 abort_connection(ep, skb, GFP_KERNEL);
938 PDBG("%s enter (%s line %u)\n", __FUNCTION__, __FILE__, __LINE__);
941 * Copy the new data into our accumulation buffer.
943 memcpy(&(ep->mpa_pkt[ep->mpa_pkt_len]), skb->data, skb->len);
944 ep->mpa_pkt_len += skb->len;
947 * If we don't even have the mpa message, then bail.
948 * We'll continue process when more data arrives.
950 if (ep->mpa_pkt_len < sizeof(*mpa))
952 PDBG("%s enter (%s line %u)\n", __FUNCTION__, __FILE__, __LINE__);
953 mpa = (struct mpa_message *) ep->mpa_pkt;
956 * Validate MPA Header.
958 if (mpa->revision != mpa_rev) {
959 abort_connection(ep, skb, GFP_KERNEL);
963 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
964 abort_connection(ep, skb, GFP_KERNEL);
968 plen = ntohs(mpa->private_data_size);
971 * Fail if there's too much private data.
973 if (plen > MPA_MAX_PRIVATE_DATA) {
974 abort_connection(ep, skb, GFP_KERNEL);
979 * If plen does not account for pkt size
981 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
982 abort_connection(ep, skb, GFP_KERNEL);
985 ep->plen = (u8) plen;
988 * If we don't have all the pdata yet, then bail.
990 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
994 * If we get here we have accumulated the entire mpa
995 * start reply message including private data.
997 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
998 ep->mpa_attr.recv_marker_enabled = markers_enabled;
999 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1000 ep->mpa_attr.version = mpa_rev;
1001 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1002 "xmit_marker_enabled=%d, version=%d\n", __FUNCTION__,
1003 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1004 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1006 state_set(&ep->com, MPA_REQ_RCVD);
1009 connect_request_upcall(ep);
1013 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1015 struct iwch_ep *ep = ctx;
1016 struct cpl_rx_data *hdr = cplhdr(skb);
1017 unsigned int dlen = ntohs(hdr->len);
1019 PDBG("%s ep %p dlen %u\n", __FUNCTION__, ep, dlen);
1021 skb_pull(skb, sizeof(*hdr));
1022 skb_trim(skb, dlen);
1024 switch (state_read(&ep->com)) {
1026 process_mpa_reply(ep, skb);
1029 process_mpa_request(ep, skb);
1034 printk(KERN_ERR MOD "%s Unexpected streaming data."
1035 " ep %p state %d tid %d\n",
1036 __FUNCTION__, ep, state_read(&ep->com), ep->hwtid);
1039 * The ep will timeout and inform the ULP of the failure.
1045 /* update RX credits */
1046 update_rx_credits(ep, dlen);
1048 return CPL_RET_BUF_DONE;
1052 * Upcall from the adapter indicating data has been transmitted.
1053 * For us its just the single MPA request or reply. We can now free
1054 * the skb holding the mpa message.
1056 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1058 struct iwch_ep *ep = ctx;
1059 struct cpl_wr_ack *hdr = cplhdr(skb);
1060 unsigned int credits = ntohs(hdr->credits);
1061 enum iwch_qp_attr_mask mask;
1063 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
1066 return CPL_RET_BUF_DONE;
1067 BUG_ON(credits != 1);
1068 BUG_ON(ep->mpa_skb == NULL);
1069 kfree_skb(ep->mpa_skb);
1071 dst_confirm(ep->dst);
1072 if (state_read(&ep->com) == MPA_REP_SENT) {
1073 struct iwch_qp_attributes attrs;
1075 /* bind QP to EP and move to RTS */
1076 attrs.mpa_attr = ep->mpa_attr;
1077 attrs.max_ird = ep->ord;
1078 attrs.max_ord = ep->ord;
1079 attrs.llp_stream_handle = ep;
1080 attrs.next_state = IWCH_QP_STATE_RTS;
1082 /* bind QP and TID with INIT_WR */
1083 mask = IWCH_QP_ATTR_NEXT_STATE |
1084 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1085 IWCH_QP_ATTR_MPA_ATTR |
1086 IWCH_QP_ATTR_MAX_IRD |
1087 IWCH_QP_ATTR_MAX_ORD;
1089 ep->com.rpl_err = iwch_modify_qp(ep->com.qp->rhp,
1090 ep->com.qp, mask, &attrs, 1);
1092 if (!ep->com.rpl_err) {
1093 state_set(&ep->com, FPDU_MODE);
1094 established_upcall(ep);
1097 ep->com.rpl_done = 1;
1098 PDBG("waking up ep %p\n", ep);
1099 wake_up(&ep->com.waitq);
1101 return CPL_RET_BUF_DONE;
1104 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1106 struct iwch_ep *ep = ctx;
1108 PDBG("%s ep %p\n", __FUNCTION__, ep);
1110 close_complete_upcall(ep);
1111 state_set(&ep->com, DEAD);
1112 release_ep_resources(ep);
1113 return CPL_RET_BUF_DONE;
1116 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1118 struct iwch_ep *ep = ctx;
1119 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1121 PDBG("%s ep %p status %u errno %d\n", __FUNCTION__, ep, rpl->status,
1122 status2errno(rpl->status));
1123 connect_reply_upcall(ep, status2errno(rpl->status));
1124 state_set(&ep->com, DEAD);
1125 if (ep->com.tdev->type == T3B)
1126 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1127 cxgb3_free_atid(ep->com.tdev, ep->atid);
1128 dst_release(ep->dst);
1129 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1131 return CPL_RET_BUF_DONE;
1134 static int listen_start(struct iwch_listen_ep *ep)
1136 struct sk_buff *skb;
1137 struct cpl_pass_open_req *req;
1139 PDBG("%s ep %p\n", __FUNCTION__, ep);
1140 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1142 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1146 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1147 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1148 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1149 req->local_port = ep->com.local_addr.sin_port;
1150 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1153 req->peer_netmask = 0;
1154 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1155 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1156 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1159 ep->com.tdev->send(ep->com.tdev, skb);
1163 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1165 struct iwch_listen_ep *ep = ctx;
1166 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1168 PDBG("%s ep %p status %d error %d\n", __FUNCTION__, ep,
1169 rpl->status, status2errno(rpl->status));
1170 ep->com.rpl_err = status2errno(rpl->status);
1171 ep->com.rpl_done = 1;
1172 wake_up(&ep->com.waitq);
1174 return CPL_RET_BUF_DONE;
1177 static int listen_stop(struct iwch_listen_ep *ep)
1179 struct sk_buff *skb;
1180 struct cpl_close_listserv_req *req;
1182 PDBG("%s ep %p\n", __FUNCTION__, ep);
1183 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1185 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
1188 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1189 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1190 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1192 ep->com.tdev->send(ep->com.tdev, skb);
1196 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1199 struct iwch_listen_ep *ep = ctx;
1200 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1202 PDBG("%s ep %p\n", __FUNCTION__, ep);
1203 ep->com.rpl_err = status2errno(rpl->status);
1204 ep->com.rpl_done = 1;
1205 wake_up(&ep->com.waitq);
1206 return CPL_RET_BUF_DONE;
1209 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1211 struct cpl_pass_accept_rpl *rpl;
1212 unsigned int mtu_idx;
1213 u32 opt0h, opt0l, opt2;
1216 PDBG("%s ep %p\n", __FUNCTION__, ep);
1217 BUG_ON(skb_cloned(skb));
1218 skb_trim(skb, sizeof(*rpl));
1220 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1221 wscale = compute_wscale(rcv_win);
1222 opt0h = V_NAGLE(0) |
1226 V_WND_SCALE(wscale) |
1227 V_MSS_IDX(mtu_idx) |
1228 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1229 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1230 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
1233 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1234 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1235 rpl->peer_ip = peer_ip;
1236 rpl->opt0h = htonl(opt0h);
1237 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1238 rpl->opt2 = htonl(opt2);
1239 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1240 skb->priority = CPL_PRIORITY_SETUP;
1241 l2t_send(ep->com.tdev, skb, ep->l2t);
1246 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1247 struct sk_buff *skb)
1249 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __FUNCTION__, tdev, hwtid,
1251 BUG_ON(skb_cloned(skb));
1252 skb_trim(skb, sizeof(struct cpl_tid_release));
1255 if (tdev->type == T3B)
1256 release_tid(tdev, hwtid, skb);
1258 struct cpl_pass_accept_rpl *rpl;
1261 skb->priority = CPL_PRIORITY_SETUP;
1262 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1263 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1265 rpl->peer_ip = peer_ip;
1266 rpl->opt0h = htonl(F_TCAM_BYPASS);
1267 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1269 rpl->rsvd = rpl->opt2;
1270 tdev->send(tdev, skb);
1274 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1276 struct iwch_ep *child_ep, *parent_ep = ctx;
1277 struct cpl_pass_accept_req *req = cplhdr(skb);
1278 unsigned int hwtid = GET_TID(req);
1279 struct dst_entry *dst;
1280 struct l2t_entry *l2t;
1284 PDBG("%s parent ep %p tid %u\n", __FUNCTION__, parent_ep, hwtid);
1286 if (state_read(&parent_ep->com) != LISTEN) {
1287 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1293 * Find the netdev for this connection request.
1295 tim.mac_addr = req->dst_mac;
1296 tim.vlan_tag = ntohs(req->vlan_tag);
1297 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1299 "%s bad dst mac %02x %02x %02x %02x %02x %02x\n",
1310 /* Find output route */
1311 rt = find_route(tdev,
1315 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1317 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1322 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1324 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1329 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1331 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1333 l2t_release(L2DATA(tdev), l2t);
1337 state_set(&child_ep->com, CONNECTING);
1338 child_ep->com.tdev = tdev;
1339 child_ep->com.cm_id = NULL;
1340 child_ep->com.local_addr.sin_family = PF_INET;
1341 child_ep->com.local_addr.sin_port = req->local_port;
1342 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1343 child_ep->com.remote_addr.sin_family = PF_INET;
1344 child_ep->com.remote_addr.sin_port = req->peer_port;
1345 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1346 get_ep(&parent_ep->com);
1347 child_ep->parent_ep = parent_ep;
1348 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1349 child_ep->l2t = l2t;
1350 child_ep->dst = dst;
1351 child_ep->hwtid = hwtid;
1352 init_timer(&child_ep->timer);
1353 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1354 accept_cr(child_ep, req->peer_ip, skb);
1357 reject_cr(tdev, hwtid, req->peer_ip, skb);
1359 return CPL_RET_BUF_DONE;
1362 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1364 struct iwch_ep *ep = ctx;
1365 struct cpl_pass_establish *req = cplhdr(skb);
1367 PDBG("%s ep %p\n", __FUNCTION__, ep);
1368 ep->snd_seq = ntohl(req->snd_isn);
1370 set_emss(ep, ntohs(req->tcp_opt));
1372 dst_confirm(ep->dst);
1373 state_set(&ep->com, MPA_REQ_WAIT);
1376 return CPL_RET_BUF_DONE;
1379 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1381 struct iwch_ep *ep = ctx;
1382 struct iwch_qp_attributes attrs;
1383 unsigned long flags;
1387 PDBG("%s ep %p\n", __FUNCTION__, ep);
1388 dst_confirm(ep->dst);
1390 spin_lock_irqsave(&ep->com.lock, flags);
1391 switch (ep->com.state) {
1393 __state_set(&ep->com, CLOSING);
1396 __state_set(&ep->com, CLOSING);
1397 connect_reply_upcall(ep, -ECONNRESET);
1402 * We're gonna mark this puppy DEAD, but keep
1403 * the reference on it until the ULP accepts or
1406 __state_set(&ep->com, CLOSING);
1410 __state_set(&ep->com, CLOSING);
1411 ep->com.rpl_done = 1;
1412 ep->com.rpl_err = -ECONNRESET;
1413 PDBG("waking up ep %p\n", ep);
1414 wake_up(&ep->com.waitq);
1418 __state_set(&ep->com, CLOSING);
1419 attrs.next_state = IWCH_QP_STATE_CLOSING;
1420 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1421 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1422 peer_close_upcall(ep);
1428 __state_set(&ep->com, MORIBUND);
1433 if (ep->com.cm_id && ep->com.qp) {
1434 attrs.next_state = IWCH_QP_STATE_IDLE;
1435 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1436 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1438 close_complete_upcall(ep);
1439 __state_set(&ep->com, DEAD);
1449 spin_unlock_irqrestore(&ep->com.lock, flags);
1451 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1453 release_ep_resources(ep);
1454 return CPL_RET_BUF_DONE;
1458 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1460 static int is_neg_adv_abort(unsigned int status)
1462 return status == CPL_ERR_RTX_NEG_ADVICE ||
1463 status == CPL_ERR_PERSIST_NEG_ADVICE;
1466 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1468 struct cpl_abort_req_rss *req = cplhdr(skb);
1469 struct iwch_ep *ep = ctx;
1470 struct cpl_abort_rpl *rpl;
1471 struct sk_buff *rpl_skb;
1472 struct iwch_qp_attributes attrs;
1476 if (is_neg_adv_abort(req->status)) {
1477 PDBG("%s neg_adv_abort ep %p tid %d\n", __FUNCTION__, ep,
1479 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1480 return CPL_RET_BUF_DONE;
1483 state = state_read(&ep->com);
1484 PDBG("%s ep %p state %u\n", __FUNCTION__, ep, state);
1493 connect_reply_upcall(ep, -ECONNRESET);
1496 ep->com.rpl_done = 1;
1497 ep->com.rpl_err = -ECONNRESET;
1498 PDBG("waking up ep %p\n", ep);
1499 wake_up(&ep->com.waitq);
1504 * We're gonna mark this puppy DEAD, but keep
1505 * the reference on it until the ULP accepts or
1515 if (ep->com.cm_id && ep->com.qp) {
1516 attrs.next_state = IWCH_QP_STATE_ERROR;
1517 ret = iwch_modify_qp(ep->com.qp->rhp,
1518 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1522 "%s - qp <- error failed!\n",
1525 peer_abort_upcall(ep);
1530 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __FUNCTION__);
1531 return CPL_RET_BUF_DONE;
1536 dst_confirm(ep->dst);
1538 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1540 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1542 dst_release(ep->dst);
1543 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1545 return CPL_RET_BUF_DONE;
1547 rpl_skb->priority = CPL_PRIORITY_DATA;
1548 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1549 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1550 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1551 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1552 rpl->cmd = CPL_ABORT_NO_RST;
1553 ep->com.tdev->send(ep->com.tdev, rpl_skb);
1554 if (state != ABORTING) {
1555 state_set(&ep->com, DEAD);
1556 release_ep_resources(ep);
1558 return CPL_RET_BUF_DONE;
1561 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1563 struct iwch_ep *ep = ctx;
1564 struct iwch_qp_attributes attrs;
1565 unsigned long flags;
1568 PDBG("%s ep %p\n", __FUNCTION__, ep);
1571 /* The cm_id may be null if we failed to connect */
1572 spin_lock_irqsave(&ep->com.lock, flags);
1573 switch (ep->com.state) {
1575 __state_set(&ep->com, MORIBUND);
1579 if ((ep->com.cm_id) && (ep->com.qp)) {
1580 attrs.next_state = IWCH_QP_STATE_IDLE;
1581 iwch_modify_qp(ep->com.qp->rhp,
1583 IWCH_QP_ATTR_NEXT_STATE,
1586 close_complete_upcall(ep);
1587 __state_set(&ep->com, DEAD);
1597 spin_unlock_irqrestore(&ep->com.lock, flags);
1599 release_ep_resources(ep);
1600 return CPL_RET_BUF_DONE;
1604 * T3A does 3 things when a TERM is received:
1605 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1606 * 2) generate an async event on the QP with the TERMINATE opcode
1607 * 3) post a TERMINATE opcde cqe into the associated CQ.
1609 * For (1), we save the message in the qp for later consumer consumption.
1610 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1611 * For (3), we toss the CQE in cxio_poll_cq().
1613 * terminate() handles case (1)...
1615 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1617 struct iwch_ep *ep = ctx;
1619 PDBG("%s ep %p\n", __FUNCTION__, ep);
1620 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1621 PDBG("%s saving %d bytes of term msg\n", __FUNCTION__, skb->len);
1622 memcpy(ep->com.qp->attr.terminate_buffer, skb->data, skb->len);
1623 ep->com.qp->attr.terminate_msg_len = skb->len;
1624 ep->com.qp->attr.is_terminate_local = 0;
1625 return CPL_RET_BUF_DONE;
1628 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1630 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1631 struct iwch_ep *ep = ctx;
1633 PDBG("%s ep %p tid %u status %d\n", __FUNCTION__, ep, ep->hwtid,
1636 struct iwch_qp_attributes attrs;
1638 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1639 __FUNCTION__, ep->hwtid);
1641 attrs.next_state = IWCH_QP_STATE_ERROR;
1642 iwch_modify_qp(ep->com.qp->rhp,
1643 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1645 abort_connection(ep, NULL, GFP_KERNEL);
1647 return CPL_RET_BUF_DONE;
1650 static void ep_timeout(unsigned long arg)
1652 struct iwch_ep *ep = (struct iwch_ep *)arg;
1653 struct iwch_qp_attributes attrs;
1654 unsigned long flags;
1656 spin_lock_irqsave(&ep->com.lock, flags);
1657 PDBG("%s ep %p tid %u state %d\n", __FUNCTION__, ep, ep->hwtid,
1659 switch (ep->com.state) {
1661 connect_reply_upcall(ep, -ETIMEDOUT);
1667 if (ep->com.cm_id && ep->com.qp) {
1668 attrs.next_state = IWCH_QP_STATE_ERROR;
1669 iwch_modify_qp(ep->com.qp->rhp,
1670 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1677 __state_set(&ep->com, CLOSING);
1678 spin_unlock_irqrestore(&ep->com.lock, flags);
1679 abort_connection(ep, NULL, GFP_ATOMIC);
1683 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1686 struct iwch_ep *ep = to_ep(cm_id);
1687 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1689 if (state_read(&ep->com) == DEAD) {
1693 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1695 abort_connection(ep, NULL, GFP_KERNEL);
1697 err = send_mpa_reject(ep, pdata, pdata_len);
1698 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1703 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1706 struct iwch_qp_attributes attrs;
1707 enum iwch_qp_attr_mask mask;
1708 struct iwch_ep *ep = to_ep(cm_id);
1709 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1710 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1712 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1713 if (state_read(&ep->com) == DEAD) {
1718 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1721 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1722 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1723 abort_connection(ep, NULL, GFP_KERNEL);
1727 cm_id->add_ref(cm_id);
1728 ep->com.cm_id = cm_id;
1731 ep->com.rpl_done = 0;
1732 ep->com.rpl_err = 0;
1733 ep->ird = conn_param->ird;
1734 ep->ord = conn_param->ord;
1735 PDBG("%s %d ird %d ord %d\n", __FUNCTION__, __LINE__, ep->ird, ep->ord);
1737 err = send_mpa_reply(ep, conn_param->private_data,
1738 conn_param->private_data_len);
1740 ep->com.cm_id = NULL;
1742 cm_id->rem_ref(cm_id);
1743 abort_connection(ep, NULL, GFP_KERNEL);
1748 /* bind QP to EP and move to RTS */
1749 attrs.mpa_attr = ep->mpa_attr;
1750 attrs.max_ird = ep->ord;
1751 attrs.max_ord = ep->ord;
1752 attrs.llp_stream_handle = ep;
1753 attrs.next_state = IWCH_QP_STATE_RTS;
1755 /* bind QP and TID with INIT_WR */
1756 mask = IWCH_QP_ATTR_NEXT_STATE |
1757 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1758 IWCH_QP_ATTR_MPA_ATTR |
1759 IWCH_QP_ATTR_MAX_IRD |
1760 IWCH_QP_ATTR_MAX_ORD;
1762 err = iwch_modify_qp(ep->com.qp->rhp,
1763 ep->com.qp, mask, &attrs, 1);
1766 ep->com.cm_id = NULL;
1768 cm_id->rem_ref(cm_id);
1769 abort_connection(ep, NULL, GFP_KERNEL);
1771 state_set(&ep->com, FPDU_MODE);
1772 established_upcall(ep);
1778 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1781 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1785 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1787 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1791 init_timer(&ep->timer);
1792 ep->plen = conn_param->private_data_len;
1794 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1795 conn_param->private_data, ep->plen);
1796 ep->ird = conn_param->ird;
1797 ep->ord = conn_param->ord;
1798 ep->com.tdev = h->rdev.t3cdev_p;
1800 cm_id->add_ref(cm_id);
1801 ep->com.cm_id = cm_id;
1802 ep->com.qp = get_qhp(h, conn_param->qpn);
1803 BUG_ON(!ep->com.qp);
1804 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __FUNCTION__, conn_param->qpn,
1808 * Allocate an active TID to initiate a TCP connection.
1810 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1811 if (ep->atid == -1) {
1812 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1818 rt = find_route(h->rdev.t3cdev_p,
1819 cm_id->local_addr.sin_addr.s_addr,
1820 cm_id->remote_addr.sin_addr.s_addr,
1821 cm_id->local_addr.sin_port,
1822 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1824 printk(KERN_ERR MOD "%s - cannot find route.\n", __FUNCTION__);
1825 err = -EHOSTUNREACH;
1828 ep->dst = &rt->u.dst;
1830 /* get a l2t entry */
1831 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1832 ep->dst->neighbour->dev);
1834 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __FUNCTION__);
1839 state_set(&ep->com, CONNECTING);
1840 ep->tos = IPTOS_LOWDELAY;
1841 ep->com.local_addr = cm_id->local_addr;
1842 ep->com.remote_addr = cm_id->remote_addr;
1844 /* send connect request to rnic */
1845 err = send_connect(ep);
1849 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1851 dst_release(ep->dst);
1853 cxgb3_free_atid(ep->com.tdev, ep->atid);
1860 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1863 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1864 struct iwch_listen_ep *ep;
1869 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1871 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1875 PDBG("%s ep %p\n", __FUNCTION__, ep);
1876 ep->com.tdev = h->rdev.t3cdev_p;
1877 cm_id->add_ref(cm_id);
1878 ep->com.cm_id = cm_id;
1879 ep->backlog = backlog;
1880 ep->com.local_addr = cm_id->local_addr;
1883 * Allocate a server TID.
1885 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1886 if (ep->stid == -1) {
1887 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1892 state_set(&ep->com, LISTEN);
1893 err = listen_start(ep);
1897 /* wait for pass_open_rpl */
1898 wait_event(ep->com.waitq, ep->com.rpl_done);
1899 err = ep->com.rpl_err;
1901 cm_id->provider_data = ep;
1905 cxgb3_free_stid(ep->com.tdev, ep->stid);
1913 int iwch_destroy_listen(struct iw_cm_id *cm_id)
1916 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
1918 PDBG("%s ep %p\n", __FUNCTION__, ep);
1921 state_set(&ep->com, DEAD);
1922 ep->com.rpl_done = 0;
1923 ep->com.rpl_err = 0;
1924 err = listen_stop(ep);
1925 wait_event(ep->com.waitq, ep->com.rpl_done);
1926 cxgb3_free_stid(ep->com.tdev, ep->stid);
1927 err = ep->com.rpl_err;
1928 cm_id->rem_ref(cm_id);
1933 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
1936 unsigned long flags;
1939 spin_lock_irqsave(&ep->com.lock, flags);
1941 PDBG("%s ep %p state %s, abrupt %d\n", __FUNCTION__, ep,
1942 states[ep->com.state], abrupt);
1944 if (ep->com.state == DEAD) {
1945 PDBG("%s already dead ep %p\n", __FUNCTION__, ep);
1950 if (ep->com.state != ABORTING) {
1951 ep->com.state = ABORTING;
1957 switch (ep->com.state) {
1964 ep->com.state = CLOSING;
1968 ep->com.state = MORIBUND;
1978 spin_unlock_irqrestore(&ep->com.lock, flags);
1981 ret = send_abort(ep, NULL, gfp);
1983 ret = send_halfclose(ep, gfp);
1988 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
1989 struct l2t_entry *l2t)
1991 struct iwch_ep *ep = ctx;
1996 PDBG("%s ep %p redirect to dst %p l2t %p\n", __FUNCTION__, ep, new,
1999 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2007 * All the CM events are handled on a work queue to have a safe context.
2009 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2011 struct iwch_ep_common *epc = ctx;
2016 * Save ctx and tdev in the skb->cb area.
2018 *((void **) skb->cb) = ctx;
2019 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2022 * Queue the skb and schedule the worker thread.
2024 skb_queue_tail(&rxq, skb);
2025 queue_work(workq, &skb_work);
2029 int __init iwch_cm_init(void)
2031 skb_queue_head_init(&rxq);
2033 workq = create_singlethread_workqueue("iw_cxgb3");
2038 * All upcalls from the T3 Core go to sched() to
2039 * schedule the processing on a work queue.
2041 t3c_handlers[CPL_ACT_ESTABLISH] = sched;
2042 t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
2043 t3c_handlers[CPL_RX_DATA] = sched;
2044 t3c_handlers[CPL_TX_DMA_ACK] = sched;
2045 t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
2046 t3c_handlers[CPL_ABORT_RPL] = sched;
2047 t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
2048 t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
2049 t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
2050 t3c_handlers[CPL_PASS_ESTABLISH] = sched;
2051 t3c_handlers[CPL_PEER_CLOSE] = sched;
2052 t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
2053 t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
2054 t3c_handlers[CPL_RDMA_TERMINATE] = sched;
2055 t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
2058 * These are the real handlers that are called from a
2061 work_handlers[CPL_ACT_ESTABLISH] = act_establish;
2062 work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
2063 work_handlers[CPL_RX_DATA] = rx_data;
2064 work_handlers[CPL_TX_DMA_ACK] = tx_ack;
2065 work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
2066 work_handlers[CPL_ABORT_RPL] = abort_rpl;
2067 work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
2068 work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
2069 work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
2070 work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
2071 work_handlers[CPL_PEER_CLOSE] = peer_close;
2072 work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
2073 work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
2074 work_handlers[CPL_RDMA_TERMINATE] = terminate;
2075 work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
2079 void __exit iwch_cm_term(void)
2081 flush_workqueue(workq);
2082 destroy_workqueue(workq);