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)
309 BUG_ON(skb_cloned(skb));
313 skb = alloc_skb(len, gfp);
318 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
319 __be32 peer_ip, __be16 local_port,
320 __be16 peer_port, u8 tos)
331 .proto = IPPROTO_TCP,
339 if (ip_route_output_flow(&rt, &fl, NULL, 0))
344 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
348 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
353 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
355 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
360 * Handle an ARP failure for an active open.
362 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
364 printk(KERN_ERR MOD "ARP failure duing connect\n");
369 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
372 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
374 struct cpl_abort_req *req = cplhdr(skb);
376 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
377 req->cmd = CPL_ABORT_NO_RST;
378 cxgb3_ofld_send(dev, skb);
381 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
383 struct cpl_close_con_req *req;
386 PDBG("%s ep %p\n", __FUNCTION__, ep);
387 skb = get_skb(NULL, sizeof(*req), gfp);
389 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
392 skb->priority = CPL_PRIORITY_DATA;
393 set_arp_failure_handler(skb, arp_failure_discard);
394 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
395 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
396 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
397 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
398 l2t_send(ep->com.tdev, skb, ep->l2t);
402 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
404 struct cpl_abort_req *req;
406 PDBG("%s ep %p\n", __FUNCTION__, ep);
407 skb = get_skb(skb, sizeof(*req), gfp);
409 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
413 skb->priority = CPL_PRIORITY_DATA;
414 set_arp_failure_handler(skb, abort_arp_failure);
415 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
416 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
417 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
418 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
419 req->cmd = CPL_ABORT_SEND_RST;
420 l2t_send(ep->com.tdev, skb, ep->l2t);
424 static int send_connect(struct iwch_ep *ep)
426 struct cpl_act_open_req *req;
428 u32 opt0h, opt0l, opt2;
429 unsigned int mtu_idx;
432 PDBG("%s ep %p\n", __FUNCTION__, ep);
434 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
436 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
440 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
441 wscale = compute_wscale(rcv_win);
446 V_WND_SCALE(wscale) |
448 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
449 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
450 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
451 skb->priority = CPL_PRIORITY_SETUP;
452 set_arp_failure_handler(skb, act_open_req_arp_failure);
454 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
455 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
456 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
457 req->local_port = ep->com.local_addr.sin_port;
458 req->peer_port = ep->com.remote_addr.sin_port;
459 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
460 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
461 req->opt0h = htonl(opt0h);
462 req->opt0l = htonl(opt0l);
464 req->opt2 = htonl(opt2);
465 l2t_send(ep->com.tdev, skb, ep->l2t);
469 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
472 struct tx_data_wr *req;
473 struct mpa_message *mpa;
476 PDBG("%s ep %p pd_len %d\n", __FUNCTION__, ep, ep->plen);
478 BUG_ON(skb_cloned(skb));
480 mpalen = sizeof(*mpa) + ep->plen;
481 if (skb->data + mpalen + sizeof(*req) > skb->end) {
483 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
485 connect_reply_upcall(ep, -ENOMEM);
490 skb_reserve(skb, sizeof(*req));
491 skb_put(skb, mpalen);
492 skb->priority = CPL_PRIORITY_DATA;
493 mpa = (struct mpa_message *) skb->data;
494 memset(mpa, 0, sizeof(*mpa));
495 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
496 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
497 (markers_enabled ? MPA_MARKERS : 0);
498 mpa->private_data_size = htons(ep->plen);
499 mpa->revision = mpa_rev;
502 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
505 * Reference the mpa skb. This ensures the data area
506 * will remain in memory until the hw acks the tx.
507 * Function tx_ack() will deref it.
510 set_arp_failure_handler(skb, arp_failure_discard);
511 skb->h.raw = skb->data;
513 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
514 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
515 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
516 req->len = htonl(len);
517 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
518 V_TX_SNDBUF(snd_win>>15));
519 req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
520 req->sndseq = htonl(ep->snd_seq);
523 l2t_send(ep->com.tdev, skb, ep->l2t);
525 state_set(&ep->com, MPA_REQ_SENT);
529 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
532 struct tx_data_wr *req;
533 struct mpa_message *mpa;
536 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
538 mpalen = sizeof(*mpa) + plen;
540 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
542 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
545 skb_reserve(skb, sizeof(*req));
546 mpa = (struct mpa_message *) skb_put(skb, mpalen);
547 memset(mpa, 0, sizeof(*mpa));
548 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
549 mpa->flags = MPA_REJECT;
550 mpa->revision = mpa_rev;
551 mpa->private_data_size = htons(plen);
553 memcpy(mpa->private_data, pdata, plen);
556 * Reference the mpa skb again. This ensures the data area
557 * will remain in memory until the hw acks the tx.
558 * Function tx_ack() will deref it.
561 skb->priority = CPL_PRIORITY_DATA;
562 set_arp_failure_handler(skb, arp_failure_discard);
563 skb->h.raw = skb->data;
564 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
565 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
566 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
567 req->len = htonl(mpalen);
568 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
569 V_TX_SNDBUF(snd_win>>15));
570 req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
571 req->sndseq = htonl(ep->snd_seq);
574 l2t_send(ep->com.tdev, skb, ep->l2t);
578 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
581 struct tx_data_wr *req;
582 struct mpa_message *mpa;
586 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
588 mpalen = sizeof(*mpa) + plen;
590 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
592 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
595 skb->priority = CPL_PRIORITY_DATA;
596 skb_reserve(skb, sizeof(*req));
597 mpa = (struct mpa_message *) skb_put(skb, mpalen);
598 memset(mpa, 0, sizeof(*mpa));
599 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
600 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
601 (markers_enabled ? MPA_MARKERS : 0);
602 mpa->revision = mpa_rev;
603 mpa->private_data_size = htons(plen);
605 memcpy(mpa->private_data, pdata, plen);
608 * Reference the mpa skb. This ensures the data area
609 * will remain in memory until the hw acks the tx.
610 * Function tx_ack() will deref it.
613 set_arp_failure_handler(skb, arp_failure_discard);
614 skb->h.raw = skb->data;
616 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
617 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
618 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
619 req->len = htonl(len);
620 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
621 V_TX_SNDBUF(snd_win>>15));
622 req->flags = htonl(F_TX_MORE | F_TX_IMM_ACK | F_TX_INIT);
623 req->sndseq = htonl(ep->snd_seq);
625 state_set(&ep->com, MPA_REP_SENT);
626 l2t_send(ep->com.tdev, skb, ep->l2t);
630 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
632 struct iwch_ep *ep = ctx;
633 struct cpl_act_establish *req = cplhdr(skb);
634 unsigned int tid = GET_TID(req);
636 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, tid);
638 dst_confirm(ep->dst);
640 /* setup the hwtid for this connection */
642 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
644 ep->snd_seq = ntohl(req->snd_isn);
646 set_emss(ep, ntohs(req->tcp_opt));
648 /* dealloc the atid */
649 cxgb3_free_atid(ep->com.tdev, ep->atid);
651 /* start MPA negotiation */
652 send_mpa_req(ep, skb);
657 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
659 PDBG("%s ep %p\n", __FILE__, ep);
660 state_set(&ep->com, ABORTING);
661 send_abort(ep, skb, gfp);
664 static void close_complete_upcall(struct iwch_ep *ep)
666 struct iw_cm_event event;
668 PDBG("%s ep %p\n", __FUNCTION__, ep);
669 memset(&event, 0, sizeof(event));
670 event.event = IW_CM_EVENT_CLOSE;
672 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
673 ep, ep->com.cm_id, ep->hwtid);
674 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
675 ep->com.cm_id->rem_ref(ep->com.cm_id);
676 ep->com.cm_id = NULL;
681 static void peer_close_upcall(struct iwch_ep *ep)
683 struct iw_cm_event event;
685 PDBG("%s ep %p\n", __FUNCTION__, ep);
686 memset(&event, 0, sizeof(event));
687 event.event = IW_CM_EVENT_DISCONNECT;
689 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
690 ep, ep->com.cm_id, ep->hwtid);
691 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
695 static void peer_abort_upcall(struct iwch_ep *ep)
697 struct iw_cm_event event;
699 PDBG("%s ep %p\n", __FUNCTION__, ep);
700 memset(&event, 0, sizeof(event));
701 event.event = IW_CM_EVENT_CLOSE;
702 event.status = -ECONNRESET;
704 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
705 ep->com.cm_id, ep->hwtid);
706 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
707 ep->com.cm_id->rem_ref(ep->com.cm_id);
708 ep->com.cm_id = NULL;
713 static void connect_reply_upcall(struct iwch_ep *ep, int status)
715 struct iw_cm_event event;
717 PDBG("%s ep %p status %d\n", __FUNCTION__, ep, status);
718 memset(&event, 0, sizeof(event));
719 event.event = IW_CM_EVENT_CONNECT_REPLY;
720 event.status = status;
721 event.local_addr = ep->com.local_addr;
722 event.remote_addr = ep->com.remote_addr;
724 if ((status == 0) || (status == -ECONNREFUSED)) {
725 event.private_data_len = ep->plen;
726 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
729 PDBG("%s ep %p tid %d status %d\n", __FUNCTION__, ep,
731 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
734 ep->com.cm_id->rem_ref(ep->com.cm_id);
735 ep->com.cm_id = NULL;
740 static void connect_request_upcall(struct iwch_ep *ep)
742 struct iw_cm_event event;
744 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
745 memset(&event, 0, sizeof(event));
746 event.event = IW_CM_EVENT_CONNECT_REQUEST;
747 event.local_addr = ep->com.local_addr;
748 event.remote_addr = ep->com.remote_addr;
749 event.private_data_len = ep->plen;
750 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
751 event.provider_data = ep;
752 if (state_read(&ep->parent_ep->com) != DEAD)
753 ep->parent_ep->com.cm_id->event_handler(
754 ep->parent_ep->com.cm_id,
756 put_ep(&ep->parent_ep->com);
757 ep->parent_ep = NULL;
760 static void established_upcall(struct iwch_ep *ep)
762 struct iw_cm_event event;
764 PDBG("%s ep %p\n", __FUNCTION__, ep);
765 memset(&event, 0, sizeof(event));
766 event.event = IW_CM_EVENT_ESTABLISHED;
768 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
769 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
773 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
775 struct cpl_rx_data_ack *req;
778 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
779 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
781 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
785 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
786 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
787 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
788 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
789 skb->priority = CPL_PRIORITY_ACK;
790 ep->com.tdev->send(ep->com.tdev, skb);
794 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
796 struct mpa_message *mpa;
798 struct iwch_qp_attributes attrs;
799 enum iwch_qp_attr_mask mask;
802 PDBG("%s ep %p\n", __FUNCTION__, ep);
805 * Stop mpa timer. If it expired, then the state has
806 * changed and we bail since ep_timeout already aborted
810 if (state_read(&ep->com) != MPA_REQ_SENT)
814 * If we get more than the supported amount of private data
815 * then we must fail this connection.
817 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
823 * copy the new data into our accumulation buffer.
825 memcpy(&(ep->mpa_pkt[ep->mpa_pkt_len]), skb->data, skb->len);
826 ep->mpa_pkt_len += skb->len;
829 * if we don't even have the mpa message, then bail.
831 if (ep->mpa_pkt_len < sizeof(*mpa))
833 mpa = (struct mpa_message *) ep->mpa_pkt;
835 /* Validate MPA header. */
836 if (mpa->revision != mpa_rev) {
840 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
845 plen = ntohs(mpa->private_data_size);
848 * Fail if there's too much private data.
850 if (plen > MPA_MAX_PRIVATE_DATA) {
856 * If plen does not account for pkt size
858 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
863 ep->plen = (u8) plen;
866 * If we don't have all the pdata yet, then bail.
867 * We'll continue process when more data arrives.
869 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
872 if (mpa->flags & MPA_REJECT) {
878 * If we get here we have accumulated the entire mpa
879 * start reply message including private data. And
880 * the MPA header is valid.
882 state_set(&ep->com, FPDU_MODE);
883 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
884 ep->mpa_attr.recv_marker_enabled = markers_enabled;
885 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
886 ep->mpa_attr.version = mpa_rev;
887 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
888 "xmit_marker_enabled=%d, version=%d\n", __FUNCTION__,
889 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
890 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
892 attrs.mpa_attr = ep->mpa_attr;
893 attrs.max_ird = ep->ird;
894 attrs.max_ord = ep->ord;
895 attrs.llp_stream_handle = ep;
896 attrs.next_state = IWCH_QP_STATE_RTS;
898 mask = IWCH_QP_ATTR_NEXT_STATE |
899 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
900 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
902 /* bind QP and TID with INIT_WR */
903 err = iwch_modify_qp(ep->com.qp->rhp,
904 ep->com.qp, mask, &attrs, 1);
908 abort_connection(ep, skb, GFP_KERNEL);
910 connect_reply_upcall(ep, err);
914 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
916 struct mpa_message *mpa;
919 PDBG("%s ep %p\n", __FUNCTION__, ep);
922 * Stop mpa timer. If it expired, then the state has
923 * changed and we bail since ep_timeout already aborted
927 if (state_read(&ep->com) != MPA_REQ_WAIT)
931 * If we get more than the supported amount of private data
932 * then we must fail this connection.
934 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
935 abort_connection(ep, skb, GFP_KERNEL);
939 PDBG("%s enter (%s line %u)\n", __FUNCTION__, __FILE__, __LINE__);
942 * Copy the new data into our accumulation buffer.
944 memcpy(&(ep->mpa_pkt[ep->mpa_pkt_len]), skb->data, skb->len);
945 ep->mpa_pkt_len += skb->len;
948 * If we don't even have the mpa message, then bail.
949 * We'll continue process when more data arrives.
951 if (ep->mpa_pkt_len < sizeof(*mpa))
953 PDBG("%s enter (%s line %u)\n", __FUNCTION__, __FILE__, __LINE__);
954 mpa = (struct mpa_message *) ep->mpa_pkt;
957 * Validate MPA Header.
959 if (mpa->revision != mpa_rev) {
960 abort_connection(ep, skb, GFP_KERNEL);
964 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
965 abort_connection(ep, skb, GFP_KERNEL);
969 plen = ntohs(mpa->private_data_size);
972 * Fail if there's too much private data.
974 if (plen > MPA_MAX_PRIVATE_DATA) {
975 abort_connection(ep, skb, GFP_KERNEL);
980 * If plen does not account for pkt size
982 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
983 abort_connection(ep, skb, GFP_KERNEL);
986 ep->plen = (u8) plen;
989 * If we don't have all the pdata yet, then bail.
991 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
995 * If we get here we have accumulated the entire mpa
996 * start reply message including private data.
998 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
999 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1000 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1001 ep->mpa_attr.version = mpa_rev;
1002 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1003 "xmit_marker_enabled=%d, version=%d\n", __FUNCTION__,
1004 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1005 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1007 state_set(&ep->com, MPA_REQ_RCVD);
1010 connect_request_upcall(ep);
1014 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1016 struct iwch_ep *ep = ctx;
1017 struct cpl_rx_data *hdr = cplhdr(skb);
1018 unsigned int dlen = ntohs(hdr->len);
1020 PDBG("%s ep %p dlen %u\n", __FUNCTION__, ep, dlen);
1022 skb_pull(skb, sizeof(*hdr));
1023 skb_trim(skb, dlen);
1025 switch (state_read(&ep->com)) {
1027 process_mpa_reply(ep, skb);
1030 process_mpa_request(ep, skb);
1035 printk(KERN_ERR MOD "%s Unexpected streaming data."
1036 " ep %p state %d tid %d\n",
1037 __FUNCTION__, ep, state_read(&ep->com), ep->hwtid);
1040 * The ep will timeout and inform the ULP of the failure.
1046 /* update RX credits */
1047 update_rx_credits(ep, dlen);
1049 return CPL_RET_BUF_DONE;
1053 * Upcall from the adapter indicating data has been transmitted.
1054 * For us its just the single MPA request or reply. We can now free
1055 * the skb holding the mpa message.
1057 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1059 struct iwch_ep *ep = ctx;
1060 struct cpl_wr_ack *hdr = cplhdr(skb);
1061 unsigned int credits = ntohs(hdr->credits);
1062 enum iwch_qp_attr_mask mask;
1064 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
1067 return CPL_RET_BUF_DONE;
1068 BUG_ON(credits != 1);
1069 BUG_ON(ep->mpa_skb == NULL);
1070 kfree_skb(ep->mpa_skb);
1072 dst_confirm(ep->dst);
1073 if (state_read(&ep->com) == MPA_REP_SENT) {
1074 struct iwch_qp_attributes attrs;
1076 /* bind QP to EP and move to RTS */
1077 attrs.mpa_attr = ep->mpa_attr;
1078 attrs.max_ird = ep->ord;
1079 attrs.max_ord = ep->ord;
1080 attrs.llp_stream_handle = ep;
1081 attrs.next_state = IWCH_QP_STATE_RTS;
1083 /* bind QP and TID with INIT_WR */
1084 mask = IWCH_QP_ATTR_NEXT_STATE |
1085 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1086 IWCH_QP_ATTR_MPA_ATTR |
1087 IWCH_QP_ATTR_MAX_IRD |
1088 IWCH_QP_ATTR_MAX_ORD;
1090 ep->com.rpl_err = iwch_modify_qp(ep->com.qp->rhp,
1091 ep->com.qp, mask, &attrs, 1);
1093 if (!ep->com.rpl_err) {
1094 state_set(&ep->com, FPDU_MODE);
1095 established_upcall(ep);
1098 ep->com.rpl_done = 1;
1099 PDBG("waking up ep %p\n", ep);
1100 wake_up(&ep->com.waitq);
1102 return CPL_RET_BUF_DONE;
1105 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1107 struct iwch_ep *ep = ctx;
1109 PDBG("%s ep %p\n", __FUNCTION__, ep);
1111 close_complete_upcall(ep);
1112 state_set(&ep->com, DEAD);
1113 release_ep_resources(ep);
1114 return CPL_RET_BUF_DONE;
1117 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1119 struct iwch_ep *ep = ctx;
1120 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1122 PDBG("%s ep %p status %u errno %d\n", __FUNCTION__, ep, rpl->status,
1123 status2errno(rpl->status));
1124 connect_reply_upcall(ep, status2errno(rpl->status));
1125 state_set(&ep->com, DEAD);
1126 if (ep->com.tdev->type == T3B)
1127 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1128 cxgb3_free_atid(ep->com.tdev, ep->atid);
1129 dst_release(ep->dst);
1130 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1132 return CPL_RET_BUF_DONE;
1135 static int listen_start(struct iwch_listen_ep *ep)
1137 struct sk_buff *skb;
1138 struct cpl_pass_open_req *req;
1140 PDBG("%s ep %p\n", __FUNCTION__, ep);
1141 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1143 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1147 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1148 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1149 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1150 req->local_port = ep->com.local_addr.sin_port;
1151 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1154 req->peer_netmask = 0;
1155 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1156 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1157 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1160 ep->com.tdev->send(ep->com.tdev, skb);
1164 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1166 struct iwch_listen_ep *ep = ctx;
1167 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1169 PDBG("%s ep %p status %d error %d\n", __FUNCTION__, ep,
1170 rpl->status, status2errno(rpl->status));
1171 ep->com.rpl_err = status2errno(rpl->status);
1172 ep->com.rpl_done = 1;
1173 wake_up(&ep->com.waitq);
1175 return CPL_RET_BUF_DONE;
1178 static int listen_stop(struct iwch_listen_ep *ep)
1180 struct sk_buff *skb;
1181 struct cpl_close_listserv_req *req;
1183 PDBG("%s ep %p\n", __FUNCTION__, ep);
1184 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1186 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
1189 req = (struct cpl_close_listserv_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_CLOSE_LISTSRV_REQ, ep->stid));
1193 ep->com.tdev->send(ep->com.tdev, skb);
1197 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1200 struct iwch_listen_ep *ep = ctx;
1201 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1203 PDBG("%s ep %p\n", __FUNCTION__, ep);
1204 ep->com.rpl_err = status2errno(rpl->status);
1205 ep->com.rpl_done = 1;
1206 wake_up(&ep->com.waitq);
1207 return CPL_RET_BUF_DONE;
1210 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1212 struct cpl_pass_accept_rpl *rpl;
1213 unsigned int mtu_idx;
1214 u32 opt0h, opt0l, opt2;
1217 PDBG("%s ep %p\n", __FUNCTION__, ep);
1218 BUG_ON(skb_cloned(skb));
1219 skb_trim(skb, sizeof(*rpl));
1221 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1222 wscale = compute_wscale(rcv_win);
1223 opt0h = V_NAGLE(0) |
1227 V_WND_SCALE(wscale) |
1228 V_MSS_IDX(mtu_idx) |
1229 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1230 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1231 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
1234 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1235 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1236 rpl->peer_ip = peer_ip;
1237 rpl->opt0h = htonl(opt0h);
1238 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1239 rpl->opt2 = htonl(opt2);
1240 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1241 skb->priority = CPL_PRIORITY_SETUP;
1242 l2t_send(ep->com.tdev, skb, ep->l2t);
1247 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1248 struct sk_buff *skb)
1250 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __FUNCTION__, tdev, hwtid,
1252 BUG_ON(skb_cloned(skb));
1253 skb_trim(skb, sizeof(struct cpl_tid_release));
1256 if (tdev->type == T3B)
1257 release_tid(tdev, hwtid, skb);
1259 struct cpl_pass_accept_rpl *rpl;
1262 skb->priority = CPL_PRIORITY_SETUP;
1263 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1264 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1266 rpl->peer_ip = peer_ip;
1267 rpl->opt0h = htonl(F_TCAM_BYPASS);
1268 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1270 rpl->rsvd = rpl->opt2;
1271 tdev->send(tdev, skb);
1275 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1277 struct iwch_ep *child_ep, *parent_ep = ctx;
1278 struct cpl_pass_accept_req *req = cplhdr(skb);
1279 unsigned int hwtid = GET_TID(req);
1280 struct dst_entry *dst;
1281 struct l2t_entry *l2t;
1285 PDBG("%s parent ep %p tid %u\n", __FUNCTION__, parent_ep, hwtid);
1287 if (state_read(&parent_ep->com) != LISTEN) {
1288 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1294 * Find the netdev for this connection request.
1296 tim.mac_addr = req->dst_mac;
1297 tim.vlan_tag = ntohs(req->vlan_tag);
1298 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1300 "%s bad dst mac %02x %02x %02x %02x %02x %02x\n",
1311 /* Find output route */
1312 rt = find_route(tdev,
1316 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1318 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1323 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1325 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1330 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1332 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1334 l2t_release(L2DATA(tdev), l2t);
1338 state_set(&child_ep->com, CONNECTING);
1339 child_ep->com.tdev = tdev;
1340 child_ep->com.cm_id = NULL;
1341 child_ep->com.local_addr.sin_family = PF_INET;
1342 child_ep->com.local_addr.sin_port = req->local_port;
1343 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1344 child_ep->com.remote_addr.sin_family = PF_INET;
1345 child_ep->com.remote_addr.sin_port = req->peer_port;
1346 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1347 get_ep(&parent_ep->com);
1348 child_ep->parent_ep = parent_ep;
1349 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1350 child_ep->l2t = l2t;
1351 child_ep->dst = dst;
1352 child_ep->hwtid = hwtid;
1353 init_timer(&child_ep->timer);
1354 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1355 accept_cr(child_ep, req->peer_ip, skb);
1358 reject_cr(tdev, hwtid, req->peer_ip, skb);
1360 return CPL_RET_BUF_DONE;
1363 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1365 struct iwch_ep *ep = ctx;
1366 struct cpl_pass_establish *req = cplhdr(skb);
1368 PDBG("%s ep %p\n", __FUNCTION__, ep);
1369 ep->snd_seq = ntohl(req->snd_isn);
1371 set_emss(ep, ntohs(req->tcp_opt));
1373 dst_confirm(ep->dst);
1374 state_set(&ep->com, MPA_REQ_WAIT);
1377 return CPL_RET_BUF_DONE;
1380 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1382 struct iwch_ep *ep = ctx;
1383 struct iwch_qp_attributes attrs;
1384 unsigned long flags;
1388 PDBG("%s ep %p\n", __FUNCTION__, ep);
1389 dst_confirm(ep->dst);
1391 spin_lock_irqsave(&ep->com.lock, flags);
1392 switch (ep->com.state) {
1394 __state_set(&ep->com, CLOSING);
1397 __state_set(&ep->com, CLOSING);
1398 connect_reply_upcall(ep, -ECONNRESET);
1403 * We're gonna mark this puppy DEAD, but keep
1404 * the reference on it until the ULP accepts or
1407 __state_set(&ep->com, CLOSING);
1411 __state_set(&ep->com, CLOSING);
1412 ep->com.rpl_done = 1;
1413 ep->com.rpl_err = -ECONNRESET;
1414 PDBG("waking up ep %p\n", ep);
1415 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);
1429 __state_set(&ep->com, MORIBUND);
1434 if (ep->com.cm_id && ep->com.qp) {
1435 attrs.next_state = IWCH_QP_STATE_IDLE;
1436 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1437 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1439 close_complete_upcall(ep);
1440 __state_set(&ep->com, DEAD);
1450 spin_unlock_irqrestore(&ep->com.lock, flags);
1452 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1454 release_ep_resources(ep);
1455 return CPL_RET_BUF_DONE;
1459 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1461 static int is_neg_adv_abort(unsigned int status)
1463 return status == CPL_ERR_RTX_NEG_ADVICE ||
1464 status == CPL_ERR_PERSIST_NEG_ADVICE;
1467 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1469 struct cpl_abort_req_rss *req = cplhdr(skb);
1470 struct iwch_ep *ep = ctx;
1471 struct cpl_abort_rpl *rpl;
1472 struct sk_buff *rpl_skb;
1473 struct iwch_qp_attributes attrs;
1477 if (is_neg_adv_abort(req->status)) {
1478 PDBG("%s neg_adv_abort ep %p tid %d\n", __FUNCTION__, ep,
1480 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1481 return CPL_RET_BUF_DONE;
1484 state = state_read(&ep->com);
1485 PDBG("%s ep %p state %u\n", __FUNCTION__, ep, state);
1492 connect_reply_upcall(ep, -ECONNRESET);
1495 ep->com.rpl_done = 1;
1496 ep->com.rpl_err = -ECONNRESET;
1497 PDBG("waking up ep %p\n", ep);
1498 wake_up(&ep->com.waitq);
1503 * We're gonna mark this puppy DEAD, but keep
1504 * the reference on it until the ULP accepts or
1513 if (ep->com.cm_id && ep->com.qp) {
1514 attrs.next_state = IWCH_QP_STATE_ERROR;
1515 ret = iwch_modify_qp(ep->com.qp->rhp,
1516 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1520 "%s - qp <- error failed!\n",
1523 peer_abort_upcall(ep);
1528 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __FUNCTION__);
1529 return CPL_RET_BUF_DONE;
1534 dst_confirm(ep->dst);
1536 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1538 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1540 dst_release(ep->dst);
1541 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1543 return CPL_RET_BUF_DONE;
1545 rpl_skb->priority = CPL_PRIORITY_DATA;
1546 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1547 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1548 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1549 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1550 rpl->cmd = CPL_ABORT_NO_RST;
1551 ep->com.tdev->send(ep->com.tdev, rpl_skb);
1552 if (state != ABORTING) {
1553 state_set(&ep->com, DEAD);
1554 release_ep_resources(ep);
1556 return CPL_RET_BUF_DONE;
1559 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1561 struct iwch_ep *ep = ctx;
1562 struct iwch_qp_attributes attrs;
1563 unsigned long flags;
1566 PDBG("%s ep %p\n", __FUNCTION__, ep);
1569 /* The cm_id may be null if we failed to connect */
1570 spin_lock_irqsave(&ep->com.lock, flags);
1571 switch (ep->com.state) {
1574 __state_set(&ep->com, MORIBUND);
1578 if ((ep->com.cm_id) && (ep->com.qp)) {
1579 attrs.next_state = IWCH_QP_STATE_IDLE;
1580 iwch_modify_qp(ep->com.qp->rhp,
1582 IWCH_QP_ATTR_NEXT_STATE,
1585 close_complete_upcall(ep);
1586 __state_set(&ep->com, DEAD);
1594 spin_unlock_irqrestore(&ep->com.lock, flags);
1596 release_ep_resources(ep);
1597 return CPL_RET_BUF_DONE;
1601 * T3A does 3 things when a TERM is received:
1602 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1603 * 2) generate an async event on the QP with the TERMINATE opcode
1604 * 3) post a TERMINATE opcde cqe into the associated CQ.
1606 * For (1), we save the message in the qp for later consumer consumption.
1607 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1608 * For (3), we toss the CQE in cxio_poll_cq().
1610 * terminate() handles case (1)...
1612 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1614 struct iwch_ep *ep = ctx;
1616 PDBG("%s ep %p\n", __FUNCTION__, ep);
1617 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1618 PDBG("%s saving %d bytes of term msg\n", __FUNCTION__, skb->len);
1619 memcpy(ep->com.qp->attr.terminate_buffer, skb->data, skb->len);
1620 ep->com.qp->attr.terminate_msg_len = skb->len;
1621 ep->com.qp->attr.is_terminate_local = 0;
1622 return CPL_RET_BUF_DONE;
1625 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1627 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1628 struct iwch_ep *ep = ctx;
1630 PDBG("%s ep %p tid %u status %d\n", __FUNCTION__, ep, ep->hwtid,
1633 struct iwch_qp_attributes attrs;
1635 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1636 __FUNCTION__, ep->hwtid);
1638 attrs.next_state = IWCH_QP_STATE_ERROR;
1639 iwch_modify_qp(ep->com.qp->rhp,
1640 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1642 abort_connection(ep, NULL, GFP_KERNEL);
1644 return CPL_RET_BUF_DONE;
1647 static void ep_timeout(unsigned long arg)
1649 struct iwch_ep *ep = (struct iwch_ep *)arg;
1650 struct iwch_qp_attributes attrs;
1651 unsigned long flags;
1653 spin_lock_irqsave(&ep->com.lock, flags);
1654 PDBG("%s ep %p tid %u state %d\n", __FUNCTION__, ep, ep->hwtid,
1656 switch (ep->com.state) {
1658 connect_reply_upcall(ep, -ETIMEDOUT);
1663 if (ep->com.cm_id && ep->com.qp) {
1664 attrs.next_state = IWCH_QP_STATE_ERROR;
1665 iwch_modify_qp(ep->com.qp->rhp,
1666 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1673 __state_set(&ep->com, CLOSING);
1674 spin_unlock_irqrestore(&ep->com.lock, flags);
1675 abort_connection(ep, NULL, GFP_ATOMIC);
1679 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1682 struct iwch_ep *ep = to_ep(cm_id);
1683 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1685 if (state_read(&ep->com) == DEAD) {
1689 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1690 state_set(&ep->com, CLOSING);
1692 abort_connection(ep, NULL, GFP_KERNEL);
1694 err = send_mpa_reject(ep, pdata, pdata_len);
1695 err = send_halfclose(ep, GFP_KERNEL);
1700 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1703 struct iwch_qp_attributes attrs;
1704 enum iwch_qp_attr_mask mask;
1705 struct iwch_ep *ep = to_ep(cm_id);
1706 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1707 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1709 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1710 if (state_read(&ep->com) == DEAD) {
1715 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1718 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1719 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1720 abort_connection(ep, NULL, GFP_KERNEL);
1724 cm_id->add_ref(cm_id);
1725 ep->com.cm_id = cm_id;
1728 ep->com.rpl_done = 0;
1729 ep->com.rpl_err = 0;
1730 ep->ird = conn_param->ird;
1731 ep->ord = conn_param->ord;
1732 PDBG("%s %d ird %d ord %d\n", __FUNCTION__, __LINE__, ep->ird, ep->ord);
1734 err = send_mpa_reply(ep, conn_param->private_data,
1735 conn_param->private_data_len);
1737 ep->com.cm_id = NULL;
1739 cm_id->rem_ref(cm_id);
1740 abort_connection(ep, NULL, GFP_KERNEL);
1745 /* bind QP to EP and move to RTS */
1746 attrs.mpa_attr = ep->mpa_attr;
1747 attrs.max_ird = ep->ord;
1748 attrs.max_ord = ep->ord;
1749 attrs.llp_stream_handle = ep;
1750 attrs.next_state = IWCH_QP_STATE_RTS;
1752 /* bind QP and TID with INIT_WR */
1753 mask = IWCH_QP_ATTR_NEXT_STATE |
1754 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1755 IWCH_QP_ATTR_MPA_ATTR |
1756 IWCH_QP_ATTR_MAX_IRD |
1757 IWCH_QP_ATTR_MAX_ORD;
1759 err = iwch_modify_qp(ep->com.qp->rhp,
1760 ep->com.qp, mask, &attrs, 1);
1763 ep->com.cm_id = NULL;
1765 cm_id->rem_ref(cm_id);
1766 abort_connection(ep, NULL, GFP_KERNEL);
1768 state_set(&ep->com, FPDU_MODE);
1769 established_upcall(ep);
1775 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1778 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1782 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1784 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1788 init_timer(&ep->timer);
1789 ep->plen = conn_param->private_data_len;
1791 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1792 conn_param->private_data, ep->plen);
1793 ep->ird = conn_param->ird;
1794 ep->ord = conn_param->ord;
1795 ep->com.tdev = h->rdev.t3cdev_p;
1797 cm_id->add_ref(cm_id);
1798 ep->com.cm_id = cm_id;
1799 ep->com.qp = get_qhp(h, conn_param->qpn);
1800 BUG_ON(!ep->com.qp);
1801 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __FUNCTION__, conn_param->qpn,
1805 * Allocate an active TID to initiate a TCP connection.
1807 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1808 if (ep->atid == -1) {
1809 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1815 rt = find_route(h->rdev.t3cdev_p,
1816 cm_id->local_addr.sin_addr.s_addr,
1817 cm_id->remote_addr.sin_addr.s_addr,
1818 cm_id->local_addr.sin_port,
1819 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1821 printk(KERN_ERR MOD "%s - cannot find route.\n", __FUNCTION__);
1822 err = -EHOSTUNREACH;
1825 ep->dst = &rt->u.dst;
1827 /* get a l2t entry */
1828 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1829 ep->dst->neighbour->dev);
1831 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __FUNCTION__);
1836 state_set(&ep->com, CONNECTING);
1837 ep->tos = IPTOS_LOWDELAY;
1838 ep->com.local_addr = cm_id->local_addr;
1839 ep->com.remote_addr = cm_id->remote_addr;
1841 /* send connect request to rnic */
1842 err = send_connect(ep);
1846 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1848 dst_release(ep->dst);
1850 cxgb3_free_atid(ep->com.tdev, ep->atid);
1857 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1860 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1861 struct iwch_listen_ep *ep;
1866 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1868 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1872 PDBG("%s ep %p\n", __FUNCTION__, ep);
1873 ep->com.tdev = h->rdev.t3cdev_p;
1874 cm_id->add_ref(cm_id);
1875 ep->com.cm_id = cm_id;
1876 ep->backlog = backlog;
1877 ep->com.local_addr = cm_id->local_addr;
1880 * Allocate a server TID.
1882 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1883 if (ep->stid == -1) {
1884 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1889 state_set(&ep->com, LISTEN);
1890 err = listen_start(ep);
1894 /* wait for pass_open_rpl */
1895 wait_event(ep->com.waitq, ep->com.rpl_done);
1896 err = ep->com.rpl_err;
1898 cm_id->provider_data = ep;
1902 cxgb3_free_stid(ep->com.tdev, ep->stid);
1910 int iwch_destroy_listen(struct iw_cm_id *cm_id)
1913 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
1915 PDBG("%s ep %p\n", __FUNCTION__, ep);
1918 state_set(&ep->com, DEAD);
1919 ep->com.rpl_done = 0;
1920 ep->com.rpl_err = 0;
1921 err = listen_stop(ep);
1922 wait_event(ep->com.waitq, ep->com.rpl_done);
1923 cxgb3_free_stid(ep->com.tdev, ep->stid);
1924 err = ep->com.rpl_err;
1925 cm_id->rem_ref(cm_id);
1930 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
1933 unsigned long flags;
1936 spin_lock_irqsave(&ep->com.lock, flags);
1938 PDBG("%s ep %p state %s, abrupt %d\n", __FUNCTION__, ep,
1939 states[ep->com.state], abrupt);
1941 if (ep->com.state == DEAD) {
1942 PDBG("%s already dead ep %p\n", __FUNCTION__, ep);
1947 if (ep->com.state != ABORTING) {
1948 ep->com.state = ABORTING;
1954 switch (ep->com.state) {
1960 ep->com.state = CLOSING;
1965 ep->com.state = MORIBUND;
1975 spin_unlock_irqrestore(&ep->com.lock, flags);
1978 ret = send_abort(ep, NULL, gfp);
1980 ret = send_halfclose(ep, gfp);
1985 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
1986 struct l2t_entry *l2t)
1988 struct iwch_ep *ep = ctx;
1993 PDBG("%s ep %p redirect to dst %p l2t %p\n", __FUNCTION__, ep, new,
1996 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2004 * All the CM events are handled on a work queue to have a safe context.
2006 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2008 struct iwch_ep_common *epc = ctx;
2013 * Save ctx and tdev in the skb->cb area.
2015 *((void **) skb->cb) = ctx;
2016 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2019 * Queue the skb and schedule the worker thread.
2021 skb_queue_tail(&rxq, skb);
2022 queue_work(workq, &skb_work);
2026 int __init iwch_cm_init(void)
2028 skb_queue_head_init(&rxq);
2030 workq = create_singlethread_workqueue("iw_cxgb3");
2035 * All upcalls from the T3 Core go to sched() to
2036 * schedule the processing on a work queue.
2038 t3c_handlers[CPL_ACT_ESTABLISH] = sched;
2039 t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
2040 t3c_handlers[CPL_RX_DATA] = sched;
2041 t3c_handlers[CPL_TX_DMA_ACK] = sched;
2042 t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
2043 t3c_handlers[CPL_ABORT_RPL] = sched;
2044 t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
2045 t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
2046 t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
2047 t3c_handlers[CPL_PASS_ESTABLISH] = sched;
2048 t3c_handlers[CPL_PEER_CLOSE] = sched;
2049 t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
2050 t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
2051 t3c_handlers[CPL_RDMA_TERMINATE] = sched;
2052 t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
2055 * These are the real handlers that are called from a
2058 work_handlers[CPL_ACT_ESTABLISH] = act_establish;
2059 work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
2060 work_handlers[CPL_RX_DATA] = rx_data;
2061 work_handlers[CPL_TX_DMA_ACK] = tx_ack;
2062 work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
2063 work_handlers[CPL_ABORT_RPL] = abort_rpl;
2064 work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
2065 work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
2066 work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
2067 work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
2068 work_handlers[CPL_PEER_CLOSE] = peer_close;
2069 work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
2070 work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
2071 work_handlers[CPL_RDMA_TERMINATE] = terminate;
2072 work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
2076 void __exit iwch_cm_term(void)
2078 flush_workqueue(workq);
2079 destroy_workqueue(workq);