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 cxgb3_ofld_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 cxgb3_ofld_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 cxgb3_ofld_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 = kzalloc(size, gfp);
234 kref_init(&epc->kref);
235 spin_lock_init(&epc->lock);
236 init_waitqueue_head(&epc->waitq);
238 PDBG("%s alloc ep %p\n", __FUNCTION__, epc);
242 void __free_ep(struct kref *kref)
244 struct iwch_ep_common *epc;
245 epc = container_of(kref, struct iwch_ep_common, kref);
246 PDBG("%s ep %p state %s\n", __FUNCTION__, epc, states[state_read(epc)]);
250 static void release_ep_resources(struct iwch_ep *ep)
252 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
253 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
254 dst_release(ep->dst);
255 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
259 static void process_work(struct work_struct *work)
261 struct sk_buff *skb = NULL;
266 while ((skb = skb_dequeue(&rxq))) {
267 ep = *((void **) (skb->cb));
268 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
269 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
270 if (ret & CPL_RET_BUF_DONE)
274 * ep was referenced in sched(), and is freed here.
276 put_ep((struct iwch_ep_common *)ep);
280 static int status2errno(int status)
285 case CPL_ERR_CONN_RESET:
287 case CPL_ERR_ARP_MISS:
288 return -EHOSTUNREACH;
289 case CPL_ERR_CONN_TIMEDOUT:
291 case CPL_ERR_TCAM_FULL:
293 case CPL_ERR_CONN_EXIST:
301 * Try and reuse skbs already allocated...
303 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
305 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
309 skb = alloc_skb(len, gfp);
314 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
315 __be32 peer_ip, __be16 local_port,
316 __be16 peer_port, u8 tos)
327 .proto = IPPROTO_TCP,
335 if (ip_route_output_flow(&rt, &fl, NULL, 0))
340 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
344 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
349 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
351 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
356 * Handle an ARP failure for an active open.
358 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
360 printk(KERN_ERR MOD "ARP failure duing connect\n");
365 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
368 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
370 struct cpl_abort_req *req = cplhdr(skb);
372 PDBG("%s t3cdev %p\n", __FUNCTION__, dev);
373 req->cmd = CPL_ABORT_NO_RST;
374 cxgb3_ofld_send(dev, skb);
377 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
379 struct cpl_close_con_req *req;
382 PDBG("%s ep %p\n", __FUNCTION__, ep);
383 skb = get_skb(NULL, sizeof(*req), gfp);
385 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
388 skb->priority = CPL_PRIORITY_DATA;
389 set_arp_failure_handler(skb, arp_failure_discard);
390 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
391 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
392 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
393 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
394 l2t_send(ep->com.tdev, skb, ep->l2t);
398 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
400 struct cpl_abort_req *req;
402 PDBG("%s ep %p\n", __FUNCTION__, ep);
403 skb = get_skb(skb, sizeof(*req), gfp);
405 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
409 skb->priority = CPL_PRIORITY_DATA;
410 set_arp_failure_handler(skb, abort_arp_failure);
411 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
412 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
413 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
414 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
415 req->cmd = CPL_ABORT_SEND_RST;
416 l2t_send(ep->com.tdev, skb, ep->l2t);
420 static int send_connect(struct iwch_ep *ep)
422 struct cpl_act_open_req *req;
424 u32 opt0h, opt0l, opt2;
425 unsigned int mtu_idx;
428 PDBG("%s ep %p\n", __FUNCTION__, ep);
430 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
432 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
436 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
437 wscale = compute_wscale(rcv_win);
442 V_WND_SCALE(wscale) |
444 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
445 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
446 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
447 skb->priority = CPL_PRIORITY_SETUP;
448 set_arp_failure_handler(skb, act_open_req_arp_failure);
450 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
451 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
452 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
453 req->local_port = ep->com.local_addr.sin_port;
454 req->peer_port = ep->com.remote_addr.sin_port;
455 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
456 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
457 req->opt0h = htonl(opt0h);
458 req->opt0l = htonl(opt0l);
460 req->opt2 = htonl(opt2);
461 l2t_send(ep->com.tdev, skb, ep->l2t);
465 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
468 struct tx_data_wr *req;
469 struct mpa_message *mpa;
472 PDBG("%s ep %p pd_len %d\n", __FUNCTION__, ep, ep->plen);
474 BUG_ON(skb_cloned(skb));
476 mpalen = sizeof(*mpa) + ep->plen;
477 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
479 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
481 connect_reply_upcall(ep, -ENOMEM);
486 skb_reserve(skb, sizeof(*req));
487 skb_put(skb, mpalen);
488 skb->priority = CPL_PRIORITY_DATA;
489 mpa = (struct mpa_message *) skb->data;
490 memset(mpa, 0, sizeof(*mpa));
491 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
492 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
493 (markers_enabled ? MPA_MARKERS : 0);
494 mpa->private_data_size = htons(ep->plen);
495 mpa->revision = mpa_rev;
498 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
501 * Reference the mpa skb. This ensures the data area
502 * will remain in memory until the hw acks the tx.
503 * Function tx_ack() will deref it.
506 set_arp_failure_handler(skb, arp_failure_discard);
507 skb_reset_transport_header(skb);
509 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
510 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
511 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
512 req->len = htonl(len);
513 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
514 V_TX_SNDBUF(snd_win>>15));
515 req->flags = htonl(F_TX_INIT);
516 req->sndseq = htonl(ep->snd_seq);
519 l2t_send(ep->com.tdev, skb, ep->l2t);
521 state_set(&ep->com, MPA_REQ_SENT);
525 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
528 struct tx_data_wr *req;
529 struct mpa_message *mpa;
532 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
534 mpalen = sizeof(*mpa) + plen;
536 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
538 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
541 skb_reserve(skb, sizeof(*req));
542 mpa = (struct mpa_message *) skb_put(skb, mpalen);
543 memset(mpa, 0, sizeof(*mpa));
544 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
545 mpa->flags = MPA_REJECT;
546 mpa->revision = mpa_rev;
547 mpa->private_data_size = htons(plen);
549 memcpy(mpa->private_data, pdata, plen);
552 * Reference the mpa skb again. This ensures the data area
553 * will remain in memory until the hw acks the tx.
554 * Function tx_ack() will deref it.
557 skb->priority = CPL_PRIORITY_DATA;
558 set_arp_failure_handler(skb, arp_failure_discard);
559 skb_reset_transport_header(skb);
560 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
561 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
562 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
563 req->len = htonl(mpalen);
564 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
565 V_TX_SNDBUF(snd_win>>15));
566 req->flags = htonl(F_TX_INIT);
567 req->sndseq = htonl(ep->snd_seq);
570 l2t_send(ep->com.tdev, skb, ep->l2t);
574 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
577 struct tx_data_wr *req;
578 struct mpa_message *mpa;
582 PDBG("%s ep %p plen %d\n", __FUNCTION__, ep, plen);
584 mpalen = sizeof(*mpa) + plen;
586 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
588 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __FUNCTION__);
591 skb->priority = CPL_PRIORITY_DATA;
592 skb_reserve(skb, sizeof(*req));
593 mpa = (struct mpa_message *) skb_put(skb, mpalen);
594 memset(mpa, 0, sizeof(*mpa));
595 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
596 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
597 (markers_enabled ? MPA_MARKERS : 0);
598 mpa->revision = mpa_rev;
599 mpa->private_data_size = htons(plen);
601 memcpy(mpa->private_data, pdata, plen);
604 * Reference the mpa skb. This ensures the data area
605 * will remain in memory until the hw acks the tx.
606 * Function tx_ack() will deref it.
609 set_arp_failure_handler(skb, arp_failure_discard);
610 skb_reset_transport_header(skb);
612 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
613 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA));
614 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
615 req->len = htonl(len);
616 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
617 V_TX_SNDBUF(snd_win>>15));
618 req->flags = htonl(F_TX_INIT);
619 req->sndseq = htonl(ep->snd_seq);
621 state_set(&ep->com, MPA_REP_SENT);
622 l2t_send(ep->com.tdev, skb, ep->l2t);
626 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
628 struct iwch_ep *ep = ctx;
629 struct cpl_act_establish *req = cplhdr(skb);
630 unsigned int tid = GET_TID(req);
632 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, tid);
634 dst_confirm(ep->dst);
636 /* setup the hwtid for this connection */
638 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
640 ep->snd_seq = ntohl(req->snd_isn);
641 ep->rcv_seq = ntohl(req->rcv_isn);
643 set_emss(ep, ntohs(req->tcp_opt));
645 /* dealloc the atid */
646 cxgb3_free_atid(ep->com.tdev, ep->atid);
648 /* start MPA negotiation */
649 send_mpa_req(ep, skb);
654 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
656 PDBG("%s ep %p\n", __FILE__, ep);
657 state_set(&ep->com, ABORTING);
658 send_abort(ep, skb, gfp);
661 static void close_complete_upcall(struct iwch_ep *ep)
663 struct iw_cm_event event;
665 PDBG("%s ep %p\n", __FUNCTION__, ep);
666 memset(&event, 0, sizeof(event));
667 event.event = IW_CM_EVENT_CLOSE;
669 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
670 ep, ep->com.cm_id, ep->hwtid);
671 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
672 ep->com.cm_id->rem_ref(ep->com.cm_id);
673 ep->com.cm_id = NULL;
678 static void peer_close_upcall(struct iwch_ep *ep)
680 struct iw_cm_event event;
682 PDBG("%s ep %p\n", __FUNCTION__, ep);
683 memset(&event, 0, sizeof(event));
684 event.event = IW_CM_EVENT_DISCONNECT;
686 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
687 ep, ep->com.cm_id, ep->hwtid);
688 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
692 static void peer_abort_upcall(struct iwch_ep *ep)
694 struct iw_cm_event event;
696 PDBG("%s ep %p\n", __FUNCTION__, ep);
697 memset(&event, 0, sizeof(event));
698 event.event = IW_CM_EVENT_CLOSE;
699 event.status = -ECONNRESET;
701 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
702 ep->com.cm_id, ep->hwtid);
703 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
704 ep->com.cm_id->rem_ref(ep->com.cm_id);
705 ep->com.cm_id = NULL;
710 static void connect_reply_upcall(struct iwch_ep *ep, int status)
712 struct iw_cm_event event;
714 PDBG("%s ep %p status %d\n", __FUNCTION__, ep, status);
715 memset(&event, 0, sizeof(event));
716 event.event = IW_CM_EVENT_CONNECT_REPLY;
717 event.status = status;
718 event.local_addr = ep->com.local_addr;
719 event.remote_addr = ep->com.remote_addr;
721 if ((status == 0) || (status == -ECONNREFUSED)) {
722 event.private_data_len = ep->plen;
723 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
726 PDBG("%s ep %p tid %d status %d\n", __FUNCTION__, ep,
728 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
731 ep->com.cm_id->rem_ref(ep->com.cm_id);
732 ep->com.cm_id = NULL;
737 static void connect_request_upcall(struct iwch_ep *ep)
739 struct iw_cm_event event;
741 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
742 memset(&event, 0, sizeof(event));
743 event.event = IW_CM_EVENT_CONNECT_REQUEST;
744 event.local_addr = ep->com.local_addr;
745 event.remote_addr = ep->com.remote_addr;
746 event.private_data_len = ep->plen;
747 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
748 event.provider_data = ep;
749 if (state_read(&ep->parent_ep->com) != DEAD)
750 ep->parent_ep->com.cm_id->event_handler(
751 ep->parent_ep->com.cm_id,
753 put_ep(&ep->parent_ep->com);
754 ep->parent_ep = NULL;
757 static void established_upcall(struct iwch_ep *ep)
759 struct iw_cm_event event;
761 PDBG("%s ep %p\n", __FUNCTION__, ep);
762 memset(&event, 0, sizeof(event));
763 event.event = IW_CM_EVENT_ESTABLISHED;
765 PDBG("%s ep %p tid %d\n", __FUNCTION__, ep, ep->hwtid);
766 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
770 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
772 struct cpl_rx_data_ack *req;
775 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
776 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
778 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
782 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
783 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
784 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
785 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
786 skb->priority = CPL_PRIORITY_ACK;
787 cxgb3_ofld_send(ep->com.tdev, skb);
791 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
793 struct mpa_message *mpa;
795 struct iwch_qp_attributes attrs;
796 enum iwch_qp_attr_mask mask;
799 PDBG("%s ep %p\n", __FUNCTION__, ep);
802 * Stop mpa timer. If it expired, then the state has
803 * changed and we bail since ep_timeout already aborted
807 if (state_read(&ep->com) != MPA_REQ_SENT)
811 * If we get more than the supported amount of private data
812 * then we must fail this connection.
814 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
820 * copy the new data into our accumulation buffer.
822 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
824 ep->mpa_pkt_len += skb->len;
827 * if we don't even have the mpa message, then bail.
829 if (ep->mpa_pkt_len < sizeof(*mpa))
831 mpa = (struct mpa_message *) ep->mpa_pkt;
833 /* Validate MPA header. */
834 if (mpa->revision != mpa_rev) {
838 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
843 plen = ntohs(mpa->private_data_size);
846 * Fail if there's too much private data.
848 if (plen > MPA_MAX_PRIVATE_DATA) {
854 * If plen does not account for pkt size
856 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
861 ep->plen = (u8) plen;
864 * If we don't have all the pdata yet, then bail.
865 * We'll continue process when more data arrives.
867 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
870 if (mpa->flags & MPA_REJECT) {
876 * If we get here we have accumulated the entire mpa
877 * start reply message including private data. And
878 * the MPA header is valid.
880 state_set(&ep->com, FPDU_MODE);
881 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
882 ep->mpa_attr.recv_marker_enabled = markers_enabled;
883 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
884 ep->mpa_attr.version = mpa_rev;
885 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
886 "xmit_marker_enabled=%d, version=%d\n", __FUNCTION__,
887 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
888 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
890 attrs.mpa_attr = ep->mpa_attr;
891 attrs.max_ird = ep->ird;
892 attrs.max_ord = ep->ord;
893 attrs.llp_stream_handle = ep;
894 attrs.next_state = IWCH_QP_STATE_RTS;
896 mask = IWCH_QP_ATTR_NEXT_STATE |
897 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
898 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
900 /* bind QP and TID with INIT_WR */
901 err = iwch_modify_qp(ep->com.qp->rhp,
902 ep->com.qp, mask, &attrs, 1);
906 abort_connection(ep, skb, GFP_KERNEL);
908 connect_reply_upcall(ep, err);
912 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
914 struct mpa_message *mpa;
917 PDBG("%s ep %p\n", __FUNCTION__, ep);
920 * Stop mpa timer. If it expired, then the state has
921 * changed and we bail since ep_timeout already aborted
925 if (state_read(&ep->com) != MPA_REQ_WAIT)
929 * If we get more than the supported amount of private data
930 * then we must fail this connection.
932 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
933 abort_connection(ep, skb, GFP_KERNEL);
937 PDBG("%s enter (%s line %u)\n", __FUNCTION__, __FILE__, __LINE__);
940 * Copy the new data into our accumulation buffer.
942 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_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 ep->rcv_seq += dlen;
1025 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1027 switch (state_read(&ep->com)) {
1029 process_mpa_reply(ep, skb);
1032 process_mpa_request(ep, skb);
1037 printk(KERN_ERR MOD "%s Unexpected streaming data."
1038 " ep %p state %d tid %d\n",
1039 __FUNCTION__, ep, state_read(&ep->com), ep->hwtid);
1042 * The ep will timeout and inform the ULP of the failure.
1048 /* update RX credits */
1049 update_rx_credits(ep, dlen);
1051 return CPL_RET_BUF_DONE;
1055 * Upcall from the adapter indicating data has been transmitted.
1056 * For us its just the single MPA request or reply. We can now free
1057 * the skb holding the mpa message.
1059 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1061 struct iwch_ep *ep = ctx;
1062 struct cpl_wr_ack *hdr = cplhdr(skb);
1063 unsigned int credits = ntohs(hdr->credits);
1065 PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
1068 return CPL_RET_BUF_DONE;
1069 BUG_ON(credits != 1);
1070 BUG_ON(ep->mpa_skb == NULL);
1071 kfree_skb(ep->mpa_skb);
1073 dst_confirm(ep->dst);
1074 if (state_read(&ep->com) == MPA_REP_SENT) {
1075 ep->com.rpl_done = 1;
1076 PDBG("waking up ep %p\n", ep);
1077 wake_up(&ep->com.waitq);
1079 return CPL_RET_BUF_DONE;
1082 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1084 struct iwch_ep *ep = ctx;
1086 PDBG("%s ep %p\n", __FUNCTION__, ep);
1089 * We get 2 abort replies from the HW. The first one must
1090 * be ignored except for scribbling that we need one more.
1092 if (!(ep->flags & ABORT_REQ_IN_PROGRESS)) {
1093 ep->flags |= ABORT_REQ_IN_PROGRESS;
1094 return CPL_RET_BUF_DONE;
1097 close_complete_upcall(ep);
1098 state_set(&ep->com, DEAD);
1099 release_ep_resources(ep);
1100 return CPL_RET_BUF_DONE;
1104 * Return whether a failed active open has allocated a TID
1106 static inline int act_open_has_tid(int status)
1108 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1109 status != CPL_ERR_ARP_MISS;
1112 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1114 struct iwch_ep *ep = ctx;
1115 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1117 PDBG("%s ep %p status %u errno %d\n", __FUNCTION__, ep, rpl->status,
1118 status2errno(rpl->status));
1119 connect_reply_upcall(ep, status2errno(rpl->status));
1120 state_set(&ep->com, DEAD);
1121 if (ep->com.tdev->type == T3B && act_open_has_tid(rpl->status))
1122 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1123 cxgb3_free_atid(ep->com.tdev, ep->atid);
1124 dst_release(ep->dst);
1125 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1127 return CPL_RET_BUF_DONE;
1130 static int listen_start(struct iwch_listen_ep *ep)
1132 struct sk_buff *skb;
1133 struct cpl_pass_open_req *req;
1135 PDBG("%s ep %p\n", __FUNCTION__, ep);
1136 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1138 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1142 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1143 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1144 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1145 req->local_port = ep->com.local_addr.sin_port;
1146 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1149 req->peer_netmask = 0;
1150 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1151 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1152 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1155 cxgb3_ofld_send(ep->com.tdev, skb);
1159 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1161 struct iwch_listen_ep *ep = ctx;
1162 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1164 PDBG("%s ep %p status %d error %d\n", __FUNCTION__, ep,
1165 rpl->status, status2errno(rpl->status));
1166 ep->com.rpl_err = status2errno(rpl->status);
1167 ep->com.rpl_done = 1;
1168 wake_up(&ep->com.waitq);
1170 return CPL_RET_BUF_DONE;
1173 static int listen_stop(struct iwch_listen_ep *ep)
1175 struct sk_buff *skb;
1176 struct cpl_close_listserv_req *req;
1178 PDBG("%s ep %p\n", __FUNCTION__, ep);
1179 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1181 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __FUNCTION__);
1184 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1185 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1187 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1189 cxgb3_ofld_send(ep->com.tdev, skb);
1193 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1196 struct iwch_listen_ep *ep = ctx;
1197 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1199 PDBG("%s ep %p\n", __FUNCTION__, ep);
1200 ep->com.rpl_err = status2errno(rpl->status);
1201 ep->com.rpl_done = 1;
1202 wake_up(&ep->com.waitq);
1203 return CPL_RET_BUF_DONE;
1206 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1208 struct cpl_pass_accept_rpl *rpl;
1209 unsigned int mtu_idx;
1210 u32 opt0h, opt0l, opt2;
1213 PDBG("%s ep %p\n", __FUNCTION__, ep);
1214 BUG_ON(skb_cloned(skb));
1215 skb_trim(skb, sizeof(*rpl));
1217 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1218 wscale = compute_wscale(rcv_win);
1219 opt0h = V_NAGLE(0) |
1223 V_WND_SCALE(wscale) |
1224 V_MSS_IDX(mtu_idx) |
1225 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1226 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1227 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
1230 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1231 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1232 rpl->peer_ip = peer_ip;
1233 rpl->opt0h = htonl(opt0h);
1234 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1235 rpl->opt2 = htonl(opt2);
1236 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1237 skb->priority = CPL_PRIORITY_SETUP;
1238 l2t_send(ep->com.tdev, skb, ep->l2t);
1243 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1244 struct sk_buff *skb)
1246 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __FUNCTION__, tdev, hwtid,
1248 BUG_ON(skb_cloned(skb));
1249 skb_trim(skb, sizeof(struct cpl_tid_release));
1252 if (tdev->type == T3B)
1253 release_tid(tdev, hwtid, skb);
1255 struct cpl_pass_accept_rpl *rpl;
1258 skb->priority = CPL_PRIORITY_SETUP;
1259 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1260 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1262 rpl->peer_ip = peer_ip;
1263 rpl->opt0h = htonl(F_TCAM_BYPASS);
1264 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1266 rpl->rsvd = rpl->opt2;
1267 cxgb3_ofld_send(tdev, skb);
1271 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1273 struct iwch_ep *child_ep, *parent_ep = ctx;
1274 struct cpl_pass_accept_req *req = cplhdr(skb);
1275 unsigned int hwtid = GET_TID(req);
1276 struct dst_entry *dst;
1277 struct l2t_entry *l2t;
1281 PDBG("%s parent ep %p tid %u\n", __FUNCTION__, parent_ep, hwtid);
1283 if (state_read(&parent_ep->com) != LISTEN) {
1284 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1290 * Find the netdev for this connection request.
1292 tim.mac_addr = req->dst_mac;
1293 tim.vlan_tag = ntohs(req->vlan_tag);
1294 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1296 "%s bad dst mac %02x %02x %02x %02x %02x %02x\n",
1307 /* Find output route */
1308 rt = find_route(tdev,
1312 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1314 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1319 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1321 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1326 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1328 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1330 l2t_release(L2DATA(tdev), l2t);
1334 state_set(&child_ep->com, CONNECTING);
1335 child_ep->com.tdev = tdev;
1336 child_ep->com.cm_id = NULL;
1337 child_ep->com.local_addr.sin_family = PF_INET;
1338 child_ep->com.local_addr.sin_port = req->local_port;
1339 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1340 child_ep->com.remote_addr.sin_family = PF_INET;
1341 child_ep->com.remote_addr.sin_port = req->peer_port;
1342 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1343 get_ep(&parent_ep->com);
1344 child_ep->parent_ep = parent_ep;
1345 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1346 child_ep->l2t = l2t;
1347 child_ep->dst = dst;
1348 child_ep->hwtid = hwtid;
1349 init_timer(&child_ep->timer);
1350 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1351 accept_cr(child_ep, req->peer_ip, skb);
1354 reject_cr(tdev, hwtid, req->peer_ip, skb);
1356 return CPL_RET_BUF_DONE;
1359 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1361 struct iwch_ep *ep = ctx;
1362 struct cpl_pass_establish *req = cplhdr(skb);
1364 PDBG("%s ep %p\n", __FUNCTION__, ep);
1365 ep->snd_seq = ntohl(req->snd_isn);
1366 ep->rcv_seq = ntohl(req->rcv_isn);
1368 set_emss(ep, ntohs(req->tcp_opt));
1370 dst_confirm(ep->dst);
1371 state_set(&ep->com, MPA_REQ_WAIT);
1374 return CPL_RET_BUF_DONE;
1377 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1379 struct iwch_ep *ep = ctx;
1380 struct iwch_qp_attributes attrs;
1381 unsigned long flags;
1385 PDBG("%s ep %p\n", __FUNCTION__, ep);
1386 dst_confirm(ep->dst);
1388 spin_lock_irqsave(&ep->com.lock, flags);
1389 switch (ep->com.state) {
1391 __state_set(&ep->com, CLOSING);
1394 __state_set(&ep->com, CLOSING);
1395 connect_reply_upcall(ep, -ECONNRESET);
1400 * We're gonna mark this puppy DEAD, but keep
1401 * the reference on it until the ULP accepts or
1404 __state_set(&ep->com, CLOSING);
1408 __state_set(&ep->com, CLOSING);
1409 ep->com.rpl_done = 1;
1410 ep->com.rpl_err = -ECONNRESET;
1411 PDBG("waking up ep %p\n", ep);
1412 wake_up(&ep->com.waitq);
1416 __state_set(&ep->com, CLOSING);
1417 attrs.next_state = IWCH_QP_STATE_CLOSING;
1418 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1419 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1420 peer_close_upcall(ep);
1426 __state_set(&ep->com, MORIBUND);
1431 if (ep->com.cm_id && ep->com.qp) {
1432 attrs.next_state = IWCH_QP_STATE_IDLE;
1433 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1434 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1436 close_complete_upcall(ep);
1437 __state_set(&ep->com, DEAD);
1447 spin_unlock_irqrestore(&ep->com.lock, flags);
1449 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1451 release_ep_resources(ep);
1452 return CPL_RET_BUF_DONE;
1456 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1458 static int is_neg_adv_abort(unsigned int status)
1460 return status == CPL_ERR_RTX_NEG_ADVICE ||
1461 status == CPL_ERR_PERSIST_NEG_ADVICE;
1464 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1466 struct cpl_abort_req_rss *req = cplhdr(skb);
1467 struct iwch_ep *ep = ctx;
1468 struct cpl_abort_rpl *rpl;
1469 struct sk_buff *rpl_skb;
1470 struct iwch_qp_attributes attrs;
1474 if (is_neg_adv_abort(req->status)) {
1475 PDBG("%s neg_adv_abort ep %p tid %d\n", __FUNCTION__, ep,
1477 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1478 return CPL_RET_BUF_DONE;
1482 * We get 2 peer aborts from the HW. The first one must
1483 * be ignored except for scribbling that we need one more.
1485 if (!(ep->flags & PEER_ABORT_IN_PROGRESS)) {
1486 ep->flags |= PEER_ABORT_IN_PROGRESS;
1487 return CPL_RET_BUF_DONE;
1490 state = state_read(&ep->com);
1491 PDBG("%s ep %p state %u\n", __FUNCTION__, ep, state);
1500 connect_reply_upcall(ep, -ECONNRESET);
1503 ep->com.rpl_done = 1;
1504 ep->com.rpl_err = -ECONNRESET;
1505 PDBG("waking up ep %p\n", ep);
1506 wake_up(&ep->com.waitq);
1511 * We're gonna mark this puppy DEAD, but keep
1512 * the reference on it until the ULP accepts or
1522 if (ep->com.cm_id && ep->com.qp) {
1523 attrs.next_state = IWCH_QP_STATE_ERROR;
1524 ret = iwch_modify_qp(ep->com.qp->rhp,
1525 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1529 "%s - qp <- error failed!\n",
1532 peer_abort_upcall(ep);
1537 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __FUNCTION__);
1538 return CPL_RET_BUF_DONE;
1543 dst_confirm(ep->dst);
1545 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1547 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1549 dst_release(ep->dst);
1550 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1552 return CPL_RET_BUF_DONE;
1554 rpl_skb->priority = CPL_PRIORITY_DATA;
1555 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1556 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1557 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1558 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1559 rpl->cmd = CPL_ABORT_NO_RST;
1560 cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1561 if (state != ABORTING) {
1562 state_set(&ep->com, DEAD);
1563 release_ep_resources(ep);
1565 return CPL_RET_BUF_DONE;
1568 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1570 struct iwch_ep *ep = ctx;
1571 struct iwch_qp_attributes attrs;
1572 unsigned long flags;
1575 PDBG("%s ep %p\n", __FUNCTION__, ep);
1578 /* The cm_id may be null if we failed to connect */
1579 spin_lock_irqsave(&ep->com.lock, flags);
1580 switch (ep->com.state) {
1582 __state_set(&ep->com, MORIBUND);
1586 if ((ep->com.cm_id) && (ep->com.qp)) {
1587 attrs.next_state = IWCH_QP_STATE_IDLE;
1588 iwch_modify_qp(ep->com.qp->rhp,
1590 IWCH_QP_ATTR_NEXT_STATE,
1593 close_complete_upcall(ep);
1594 __state_set(&ep->com, DEAD);
1604 spin_unlock_irqrestore(&ep->com.lock, flags);
1606 release_ep_resources(ep);
1607 return CPL_RET_BUF_DONE;
1611 * T3A does 3 things when a TERM is received:
1612 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1613 * 2) generate an async event on the QP with the TERMINATE opcode
1614 * 3) post a TERMINATE opcde cqe into the associated CQ.
1616 * For (1), we save the message in the qp for later consumer consumption.
1617 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1618 * For (3), we toss the CQE in cxio_poll_cq().
1620 * terminate() handles case (1)...
1622 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1624 struct iwch_ep *ep = ctx;
1626 PDBG("%s ep %p\n", __FUNCTION__, ep);
1627 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1628 PDBG("%s saving %d bytes of term msg\n", __FUNCTION__, skb->len);
1629 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1631 ep->com.qp->attr.terminate_msg_len = skb->len;
1632 ep->com.qp->attr.is_terminate_local = 0;
1633 return CPL_RET_BUF_DONE;
1636 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1638 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1639 struct iwch_ep *ep = ctx;
1641 PDBG("%s ep %p tid %u status %d\n", __FUNCTION__, ep, ep->hwtid,
1644 struct iwch_qp_attributes attrs;
1646 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1647 __FUNCTION__, ep->hwtid);
1649 attrs.next_state = IWCH_QP_STATE_ERROR;
1650 iwch_modify_qp(ep->com.qp->rhp,
1651 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1653 abort_connection(ep, NULL, GFP_KERNEL);
1655 return CPL_RET_BUF_DONE;
1658 static void ep_timeout(unsigned long arg)
1660 struct iwch_ep *ep = (struct iwch_ep *)arg;
1661 struct iwch_qp_attributes attrs;
1662 unsigned long flags;
1664 spin_lock_irqsave(&ep->com.lock, flags);
1665 PDBG("%s ep %p tid %u state %d\n", __FUNCTION__, ep, ep->hwtid,
1667 switch (ep->com.state) {
1669 connect_reply_upcall(ep, -ETIMEDOUT);
1675 if (ep->com.cm_id && ep->com.qp) {
1676 attrs.next_state = IWCH_QP_STATE_ERROR;
1677 iwch_modify_qp(ep->com.qp->rhp,
1678 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1685 __state_set(&ep->com, CLOSING);
1686 spin_unlock_irqrestore(&ep->com.lock, flags);
1687 abort_connection(ep, NULL, GFP_ATOMIC);
1691 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1694 struct iwch_ep *ep = to_ep(cm_id);
1695 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1697 if (state_read(&ep->com) == DEAD) {
1701 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1703 abort_connection(ep, NULL, GFP_KERNEL);
1705 err = send_mpa_reject(ep, pdata, pdata_len);
1706 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1711 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1714 struct iwch_qp_attributes attrs;
1715 enum iwch_qp_attr_mask mask;
1716 struct iwch_ep *ep = to_ep(cm_id);
1717 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1718 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1720 PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
1721 if (state_read(&ep->com) == DEAD)
1724 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1727 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1728 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1729 abort_connection(ep, NULL, GFP_KERNEL);
1733 cm_id->add_ref(cm_id);
1734 ep->com.cm_id = cm_id;
1737 ep->com.rpl_done = 0;
1738 ep->com.rpl_err = 0;
1739 ep->ird = conn_param->ird;
1740 ep->ord = conn_param->ord;
1741 PDBG("%s %d ird %d ord %d\n", __FUNCTION__, __LINE__, ep->ird, ep->ord);
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);
1764 err = send_mpa_reply(ep, conn_param->private_data,
1765 conn_param->private_data_len);
1769 /* wait for wr_ack */
1770 wait_event(ep->com.waitq, ep->com.rpl_done);
1771 err = ep->com.rpl_err;
1775 state_set(&ep->com, FPDU_MODE);
1776 established_upcall(ep);
1780 ep->com.cm_id = NULL;
1782 cm_id->rem_ref(cm_id);
1787 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1790 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1794 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1796 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1800 init_timer(&ep->timer);
1801 ep->plen = conn_param->private_data_len;
1803 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1804 conn_param->private_data, ep->plen);
1805 ep->ird = conn_param->ird;
1806 ep->ord = conn_param->ord;
1807 ep->com.tdev = h->rdev.t3cdev_p;
1809 cm_id->add_ref(cm_id);
1810 ep->com.cm_id = cm_id;
1811 ep->com.qp = get_qhp(h, conn_param->qpn);
1812 BUG_ON(!ep->com.qp);
1813 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __FUNCTION__, conn_param->qpn,
1817 * Allocate an active TID to initiate a TCP connection.
1819 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1820 if (ep->atid == -1) {
1821 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1827 rt = find_route(h->rdev.t3cdev_p,
1828 cm_id->local_addr.sin_addr.s_addr,
1829 cm_id->remote_addr.sin_addr.s_addr,
1830 cm_id->local_addr.sin_port,
1831 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1833 printk(KERN_ERR MOD "%s - cannot find route.\n", __FUNCTION__);
1834 err = -EHOSTUNREACH;
1837 ep->dst = &rt->u.dst;
1839 /* get a l2t entry */
1840 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1841 ep->dst->neighbour->dev);
1843 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __FUNCTION__);
1848 state_set(&ep->com, CONNECTING);
1849 ep->tos = IPTOS_LOWDELAY;
1850 ep->com.local_addr = cm_id->local_addr;
1851 ep->com.remote_addr = cm_id->remote_addr;
1853 /* send connect request to rnic */
1854 err = send_connect(ep);
1858 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1860 dst_release(ep->dst);
1862 cxgb3_free_atid(ep->com.tdev, ep->atid);
1869 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1872 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1873 struct iwch_listen_ep *ep;
1878 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1880 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __FUNCTION__);
1884 PDBG("%s ep %p\n", __FUNCTION__, ep);
1885 ep->com.tdev = h->rdev.t3cdev_p;
1886 cm_id->add_ref(cm_id);
1887 ep->com.cm_id = cm_id;
1888 ep->backlog = backlog;
1889 ep->com.local_addr = cm_id->local_addr;
1892 * Allocate a server TID.
1894 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1895 if (ep->stid == -1) {
1896 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __FUNCTION__);
1901 state_set(&ep->com, LISTEN);
1902 err = listen_start(ep);
1906 /* wait for pass_open_rpl */
1907 wait_event(ep->com.waitq, ep->com.rpl_done);
1908 err = ep->com.rpl_err;
1910 cm_id->provider_data = ep;
1914 cxgb3_free_stid(ep->com.tdev, ep->stid);
1916 cm_id->rem_ref(cm_id);
1923 int iwch_destroy_listen(struct iw_cm_id *cm_id)
1926 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
1928 PDBG("%s ep %p\n", __FUNCTION__, ep);
1931 state_set(&ep->com, DEAD);
1932 ep->com.rpl_done = 0;
1933 ep->com.rpl_err = 0;
1934 err = listen_stop(ep);
1935 wait_event(ep->com.waitq, ep->com.rpl_done);
1936 cxgb3_free_stid(ep->com.tdev, ep->stid);
1937 err = ep->com.rpl_err;
1938 cm_id->rem_ref(cm_id);
1943 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
1946 unsigned long flags;
1949 spin_lock_irqsave(&ep->com.lock, flags);
1951 PDBG("%s ep %p state %s, abrupt %d\n", __FUNCTION__, ep,
1952 states[ep->com.state], abrupt);
1954 if (ep->com.state == DEAD) {
1955 PDBG("%s already dead ep %p\n", __FUNCTION__, ep);
1960 if (ep->com.state != ABORTING) {
1961 ep->com.state = ABORTING;
1967 switch (ep->com.state) {
1974 ep->com.state = CLOSING;
1978 ep->com.state = MORIBUND;
1988 spin_unlock_irqrestore(&ep->com.lock, flags);
1991 ret = send_abort(ep, NULL, gfp);
1993 ret = send_halfclose(ep, gfp);
1998 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
1999 struct l2t_entry *l2t)
2001 struct iwch_ep *ep = ctx;
2006 PDBG("%s ep %p redirect to dst %p l2t %p\n", __FUNCTION__, ep, new,
2009 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2017 * All the CM events are handled on a work queue to have a safe context.
2019 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2021 struct iwch_ep_common *epc = ctx;
2026 * Save ctx and tdev in the skb->cb area.
2028 *((void **) skb->cb) = ctx;
2029 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2032 * Queue the skb and schedule the worker thread.
2034 skb_queue_tail(&rxq, skb);
2035 queue_work(workq, &skb_work);
2039 static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2041 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2043 if (rpl->status != CPL_ERR_NONE) {
2044 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2045 "for tid %u\n", rpl->status, GET_TID(rpl));
2047 return CPL_RET_BUF_DONE;
2050 int __init iwch_cm_init(void)
2052 skb_queue_head_init(&rxq);
2054 workq = create_singlethread_workqueue("iw_cxgb3");
2059 * All upcalls from the T3 Core go to sched() to
2060 * schedule the processing on a work queue.
2062 t3c_handlers[CPL_ACT_ESTABLISH] = sched;
2063 t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
2064 t3c_handlers[CPL_RX_DATA] = sched;
2065 t3c_handlers[CPL_TX_DMA_ACK] = sched;
2066 t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
2067 t3c_handlers[CPL_ABORT_RPL] = sched;
2068 t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
2069 t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
2070 t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
2071 t3c_handlers[CPL_PASS_ESTABLISH] = sched;
2072 t3c_handlers[CPL_PEER_CLOSE] = sched;
2073 t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
2074 t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
2075 t3c_handlers[CPL_RDMA_TERMINATE] = sched;
2076 t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
2077 t3c_handlers[CPL_SET_TCB_RPL] = set_tcb_rpl;
2080 * These are the real handlers that are called from a
2083 work_handlers[CPL_ACT_ESTABLISH] = act_establish;
2084 work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
2085 work_handlers[CPL_RX_DATA] = rx_data;
2086 work_handlers[CPL_TX_DMA_ACK] = tx_ack;
2087 work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
2088 work_handlers[CPL_ABORT_RPL] = abort_rpl;
2089 work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
2090 work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
2091 work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
2092 work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
2093 work_handlers[CPL_PEER_CLOSE] = peer_close;
2094 work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
2095 work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
2096 work_handlers[CPL_RDMA_TERMINATE] = terminate;
2097 work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
2101 void __exit iwch_cm_term(void)
2103 flush_workqueue(workq);
2104 destroy_workqueue(workq);