2 * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/slab.h>
35 #include <linux/workqueue.h>
36 #include <linux/skbuff.h>
37 #include <linux/timer.h>
38 #include <linux/notifier.h>
39 #include <linux/inetdevice.h>
41 #include <net/neighbour.h>
42 #include <net/netevent.h>
43 #include <net/route.h>
46 #include "cxgb3_offload.h"
48 #include "iwch_provider.h"
51 static char *states[] = {
68 module_param(peer2peer, int, 0644);
69 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
71 static int ep_timeout_secs = 60;
72 module_param(ep_timeout_secs, int, 0644);
73 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
74 "in seconds (default=60)");
76 static int mpa_rev = 1;
77 module_param(mpa_rev, int, 0644);
78 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
79 "1 is spec compliant. (default=1)");
81 static int markers_enabled = 0;
82 module_param(markers_enabled, int, 0644);
83 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
85 static int crc_enabled = 1;
86 module_param(crc_enabled, int, 0644);
87 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
89 static int rcv_win = 256 * 1024;
90 module_param(rcv_win, int, 0644);
91 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
93 static int snd_win = 32 * 1024;
94 module_param(snd_win, int, 0644);
95 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
97 static unsigned int nocong = 0;
98 module_param(nocong, uint, 0644);
99 MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
101 static unsigned int cong_flavor = 1;
102 module_param(cong_flavor, uint, 0644);
103 MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
105 static struct workqueue_struct *workq;
107 static struct sk_buff_head rxq;
109 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
110 static void ep_timeout(unsigned long arg);
111 static void connect_reply_upcall(struct iwch_ep *ep, int status);
113 static void start_ep_timer(struct iwch_ep *ep)
115 PDBG("%s ep %p\n", __func__, ep);
116 if (timer_pending(&ep->timer)) {
117 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
118 del_timer_sync(&ep->timer);
121 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
122 ep->timer.data = (unsigned long)ep;
123 ep->timer.function = ep_timeout;
124 add_timer(&ep->timer);
127 static void stop_ep_timer(struct iwch_ep *ep)
129 PDBG("%s ep %p\n", __func__, ep);
130 if (!timer_pending(&ep->timer)) {
131 printk(KERN_ERR "%s timer stopped when its not running! ep %p state %u\n",
132 __func__, ep, ep->com.state);
136 del_timer_sync(&ep->timer);
140 static int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
143 struct cxio_rdev *rdev;
145 rdev = (struct cxio_rdev *)tdev->ulp;
146 if (cxio_fatal_error(rdev)) {
150 error = l2t_send(tdev, skb, l2e);
156 int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
159 struct cxio_rdev *rdev;
161 rdev = (struct cxio_rdev *)tdev->ulp;
162 if (cxio_fatal_error(rdev)) {
166 error = cxgb3_ofld_send(tdev, skb);
172 static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
174 struct cpl_tid_release *req;
176 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
179 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
180 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
181 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
182 skb->priority = CPL_PRIORITY_SETUP;
183 iwch_cxgb3_ofld_send(tdev, skb);
187 int iwch_quiesce_tid(struct iwch_ep *ep)
189 struct cpl_set_tcb_field *req;
190 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
194 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
195 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
196 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
197 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
200 req->word = htons(W_TCB_RX_QUIESCE);
201 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
202 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
204 skb->priority = CPL_PRIORITY_DATA;
205 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
208 int iwch_resume_tid(struct iwch_ep *ep)
210 struct cpl_set_tcb_field *req;
211 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
215 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
216 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
217 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
218 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
221 req->word = htons(W_TCB_RX_QUIESCE);
222 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
225 skb->priority = CPL_PRIORITY_DATA;
226 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
229 static void set_emss(struct iwch_ep *ep, u16 opt)
231 PDBG("%s ep %p opt %u\n", __func__, ep, opt);
232 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
233 if (G_TCPOPT_TSTAMP(opt))
237 PDBG("emss=%d\n", ep->emss);
240 static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
243 enum iwch_ep_state state;
245 spin_lock_irqsave(&epc->lock, flags);
247 spin_unlock_irqrestore(&epc->lock, flags);
251 static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
256 static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
260 spin_lock_irqsave(&epc->lock, flags);
261 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
262 __state_set(epc, new);
263 spin_unlock_irqrestore(&epc->lock, flags);
267 static void *alloc_ep(int size, gfp_t gfp)
269 struct iwch_ep_common *epc;
271 epc = kzalloc(size, gfp);
273 kref_init(&epc->kref);
274 spin_lock_init(&epc->lock);
275 init_waitqueue_head(&epc->waitq);
277 PDBG("%s alloc ep %p\n", __func__, epc);
281 void __free_ep(struct kref *kref)
284 ep = container_of(container_of(kref, struct iwch_ep_common, kref),
285 struct iwch_ep, com);
286 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
287 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
288 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
289 dst_release(ep->dst);
290 l2t_release(ep->com.tdev, ep->l2t);
295 static void release_ep_resources(struct iwch_ep *ep)
297 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
298 set_bit(RELEASE_RESOURCES, &ep->com.flags);
302 static int status2errno(int status)
307 case CPL_ERR_CONN_RESET:
309 case CPL_ERR_ARP_MISS:
310 return -EHOSTUNREACH;
311 case CPL_ERR_CONN_TIMEDOUT:
313 case CPL_ERR_TCAM_FULL:
315 case CPL_ERR_CONN_EXIST:
323 * Try and reuse skbs already allocated...
325 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
327 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
331 skb = alloc_skb(len, gfp);
336 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
337 __be32 peer_ip, __be16 local_port,
338 __be16 peer_port, u8 tos)
343 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
344 peer_port, local_port, IPPROTO_TCP,
351 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
355 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
360 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
362 PDBG("%s t3cdev %p\n", __func__, dev);
367 * Handle an ARP failure for an active open.
369 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
371 printk(KERN_ERR MOD "ARP failure duing connect\n");
376 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
379 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
381 struct cpl_abort_req *req = cplhdr(skb);
383 PDBG("%s t3cdev %p\n", __func__, dev);
384 req->cmd = CPL_ABORT_NO_RST;
385 iwch_cxgb3_ofld_send(dev, skb);
388 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
390 struct cpl_close_con_req *req;
393 PDBG("%s ep %p\n", __func__, ep);
394 skb = get_skb(NULL, sizeof(*req), gfp);
396 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
399 skb->priority = CPL_PRIORITY_DATA;
400 set_arp_failure_handler(skb, arp_failure_discard);
401 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
402 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
403 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
404 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
405 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
408 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
410 struct cpl_abort_req *req;
412 PDBG("%s ep %p\n", __func__, ep);
413 skb = get_skb(skb, sizeof(*req), gfp);
415 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
419 skb->priority = CPL_PRIORITY_DATA;
420 set_arp_failure_handler(skb, abort_arp_failure);
421 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
422 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
423 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
424 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
425 req->cmd = CPL_ABORT_SEND_RST;
426 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
429 static int send_connect(struct iwch_ep *ep)
431 struct cpl_act_open_req *req;
433 u32 opt0h, opt0l, opt2;
434 unsigned int mtu_idx;
437 PDBG("%s ep %p\n", __func__, ep);
439 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
441 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
445 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
446 wscale = compute_wscale(rcv_win);
451 V_WND_SCALE(wscale) |
453 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
454 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
455 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
456 V_CONG_CONTROL_FLAVOR(cong_flavor);
457 skb->priority = CPL_PRIORITY_SETUP;
458 set_arp_failure_handler(skb, act_open_req_arp_failure);
460 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
461 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
462 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
463 req->local_port = ep->com.local_addr.sin_port;
464 req->peer_port = ep->com.remote_addr.sin_port;
465 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
466 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
467 req->opt0h = htonl(opt0h);
468 req->opt0l = htonl(opt0l);
470 req->opt2 = htonl(opt2);
471 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
474 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
477 struct tx_data_wr *req;
478 struct mpa_message *mpa;
481 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
483 BUG_ON(skb_cloned(skb));
485 mpalen = sizeof(*mpa) + ep->plen;
486 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
488 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
490 connect_reply_upcall(ep, -ENOMEM);
495 skb_reserve(skb, sizeof(*req));
496 skb_put(skb, mpalen);
497 skb->priority = CPL_PRIORITY_DATA;
498 mpa = (struct mpa_message *) skb->data;
499 memset(mpa, 0, sizeof(*mpa));
500 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
501 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
502 (markers_enabled ? MPA_MARKERS : 0);
503 mpa->private_data_size = htons(ep->plen);
504 mpa->revision = mpa_rev;
507 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
510 * Reference the mpa skb. This ensures the data area
511 * will remain in memory until the hw acks the tx.
512 * Function tx_ack() will deref it.
515 set_arp_failure_handler(skb, arp_failure_discard);
516 skb_reset_transport_header(skb);
518 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
519 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
520 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
521 req->len = htonl(len);
522 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
523 V_TX_SNDBUF(snd_win>>15));
524 req->flags = htonl(F_TX_INIT);
525 req->sndseq = htonl(ep->snd_seq);
528 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
530 state_set(&ep->com, MPA_REQ_SENT);
534 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
537 struct tx_data_wr *req;
538 struct mpa_message *mpa;
541 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
543 mpalen = sizeof(*mpa) + plen;
545 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
547 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
550 skb_reserve(skb, sizeof(*req));
551 mpa = (struct mpa_message *) skb_put(skb, mpalen);
552 memset(mpa, 0, sizeof(*mpa));
553 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
554 mpa->flags = MPA_REJECT;
555 mpa->revision = mpa_rev;
556 mpa->private_data_size = htons(plen);
558 memcpy(mpa->private_data, pdata, plen);
561 * Reference the mpa skb again. This ensures the data area
562 * will remain in memory until the hw acks the tx.
563 * Function tx_ack() will deref it.
566 skb->priority = CPL_PRIORITY_DATA;
567 set_arp_failure_handler(skb, arp_failure_discard);
568 skb_reset_transport_header(skb);
569 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
570 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
571 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
572 req->len = htonl(mpalen);
573 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
574 V_TX_SNDBUF(snd_win>>15));
575 req->flags = htonl(F_TX_INIT);
576 req->sndseq = htonl(ep->snd_seq);
579 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
582 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
585 struct tx_data_wr *req;
586 struct mpa_message *mpa;
590 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
592 mpalen = sizeof(*mpa) + plen;
594 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
596 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
599 skb->priority = CPL_PRIORITY_DATA;
600 skb_reserve(skb, sizeof(*req));
601 mpa = (struct mpa_message *) skb_put(skb, mpalen);
602 memset(mpa, 0, sizeof(*mpa));
603 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
604 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
605 (markers_enabled ? MPA_MARKERS : 0);
606 mpa->revision = mpa_rev;
607 mpa->private_data_size = htons(plen);
609 memcpy(mpa->private_data, pdata, plen);
612 * Reference the mpa skb. This ensures the data area
613 * will remain in memory until the hw acks the tx.
614 * Function tx_ack() will deref it.
617 set_arp_failure_handler(skb, arp_failure_discard);
618 skb_reset_transport_header(skb);
620 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
621 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
622 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
623 req->len = htonl(len);
624 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
625 V_TX_SNDBUF(snd_win>>15));
626 req->flags = htonl(F_TX_INIT);
627 req->sndseq = htonl(ep->snd_seq);
629 state_set(&ep->com, MPA_REP_SENT);
630 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
633 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
635 struct iwch_ep *ep = ctx;
636 struct cpl_act_establish *req = cplhdr(skb);
637 unsigned int tid = GET_TID(req);
639 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
641 dst_confirm(ep->dst);
643 /* setup the hwtid for this connection */
645 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
647 ep->snd_seq = ntohl(req->snd_isn);
648 ep->rcv_seq = ntohl(req->rcv_isn);
650 set_emss(ep, ntohs(req->tcp_opt));
652 /* dealloc the atid */
653 cxgb3_free_atid(ep->com.tdev, ep->atid);
655 /* start MPA negotiation */
656 send_mpa_req(ep, skb);
661 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
663 PDBG("%s ep %p\n", __FILE__, ep);
664 state_set(&ep->com, ABORTING);
665 send_abort(ep, skb, gfp);
668 static void close_complete_upcall(struct iwch_ep *ep)
670 struct iw_cm_event event;
672 PDBG("%s ep %p\n", __func__, ep);
673 memset(&event, 0, sizeof(event));
674 event.event = IW_CM_EVENT_CLOSE;
676 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
677 ep, ep->com.cm_id, ep->hwtid);
678 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
679 ep->com.cm_id->rem_ref(ep->com.cm_id);
680 ep->com.cm_id = NULL;
685 static void peer_close_upcall(struct iwch_ep *ep)
687 struct iw_cm_event event;
689 PDBG("%s ep %p\n", __func__, ep);
690 memset(&event, 0, sizeof(event));
691 event.event = IW_CM_EVENT_DISCONNECT;
693 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
694 ep, ep->com.cm_id, ep->hwtid);
695 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
699 static void peer_abort_upcall(struct iwch_ep *ep)
701 struct iw_cm_event event;
703 PDBG("%s ep %p\n", __func__, ep);
704 memset(&event, 0, sizeof(event));
705 event.event = IW_CM_EVENT_CLOSE;
706 event.status = -ECONNRESET;
708 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
709 ep->com.cm_id, ep->hwtid);
710 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
711 ep->com.cm_id->rem_ref(ep->com.cm_id);
712 ep->com.cm_id = NULL;
717 static void connect_reply_upcall(struct iwch_ep *ep, int status)
719 struct iw_cm_event event;
721 PDBG("%s ep %p status %d\n", __func__, ep, status);
722 memset(&event, 0, sizeof(event));
723 event.event = IW_CM_EVENT_CONNECT_REPLY;
724 event.status = status;
725 event.local_addr = ep->com.local_addr;
726 event.remote_addr = ep->com.remote_addr;
728 if ((status == 0) || (status == -ECONNREFUSED)) {
729 event.private_data_len = ep->plen;
730 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
733 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
735 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
738 ep->com.cm_id->rem_ref(ep->com.cm_id);
739 ep->com.cm_id = NULL;
744 static void connect_request_upcall(struct iwch_ep *ep)
746 struct iw_cm_event event;
748 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
749 memset(&event, 0, sizeof(event));
750 event.event = IW_CM_EVENT_CONNECT_REQUEST;
751 event.local_addr = ep->com.local_addr;
752 event.remote_addr = ep->com.remote_addr;
753 event.private_data_len = ep->plen;
754 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
755 event.provider_data = ep;
757 * Until ird/ord negotiation via MPAv2 support is added, send max
760 event.ird = event.ord = 8;
761 if (state_read(&ep->parent_ep->com) != DEAD) {
763 ep->parent_ep->com.cm_id->event_handler(
764 ep->parent_ep->com.cm_id,
767 put_ep(&ep->parent_ep->com);
768 ep->parent_ep = NULL;
771 static void established_upcall(struct iwch_ep *ep)
773 struct iw_cm_event event;
775 PDBG("%s ep %p\n", __func__, ep);
776 memset(&event, 0, sizeof(event));
777 event.event = IW_CM_EVENT_ESTABLISHED;
779 * Until ird/ord negotiation via MPAv2 support is added, send max
782 event.ird = event.ord = 8;
784 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
785 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
789 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
791 struct cpl_rx_data_ack *req;
794 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
795 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
797 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
801 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
802 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
803 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
804 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
805 skb->priority = CPL_PRIORITY_ACK;
806 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
810 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
812 struct mpa_message *mpa;
814 struct iwch_qp_attributes attrs;
815 enum iwch_qp_attr_mask mask;
818 PDBG("%s ep %p\n", __func__, ep);
821 * Stop mpa timer. If it expired, then the state has
822 * changed and we bail since ep_timeout already aborted
826 if (state_read(&ep->com) != MPA_REQ_SENT)
830 * If we get more than the supported amount of private data
831 * then we must fail this connection.
833 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
839 * copy the new data into our accumulation buffer.
841 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
843 ep->mpa_pkt_len += skb->len;
846 * if we don't even have the mpa message, then bail.
848 if (ep->mpa_pkt_len < sizeof(*mpa))
850 mpa = (struct mpa_message *) ep->mpa_pkt;
852 /* Validate MPA header. */
853 if (mpa->revision != mpa_rev) {
857 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
862 plen = ntohs(mpa->private_data_size);
865 * Fail if there's too much private data.
867 if (plen > MPA_MAX_PRIVATE_DATA) {
873 * If plen does not account for pkt size
875 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
880 ep->plen = (u8) plen;
883 * If we don't have all the pdata yet, then bail.
884 * We'll continue process when more data arrives.
886 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
889 if (mpa->flags & MPA_REJECT) {
895 * If we get here we have accumulated the entire mpa
896 * start reply message including private data. And
897 * the MPA header is valid.
899 state_set(&ep->com, FPDU_MODE);
900 ep->mpa_attr.initiator = 1;
901 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
902 ep->mpa_attr.recv_marker_enabled = markers_enabled;
903 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
904 ep->mpa_attr.version = mpa_rev;
905 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
906 "xmit_marker_enabled=%d, version=%d\n", __func__,
907 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
908 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
910 attrs.mpa_attr = ep->mpa_attr;
911 attrs.max_ird = ep->ird;
912 attrs.max_ord = ep->ord;
913 attrs.llp_stream_handle = ep;
914 attrs.next_state = IWCH_QP_STATE_RTS;
916 mask = IWCH_QP_ATTR_NEXT_STATE |
917 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
918 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
920 /* bind QP and TID with INIT_WR */
921 err = iwch_modify_qp(ep->com.qp->rhp,
922 ep->com.qp, mask, &attrs, 1);
926 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
927 iwch_post_zb_read(ep);
932 abort_connection(ep, skb, GFP_KERNEL);
934 connect_reply_upcall(ep, err);
938 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
940 struct mpa_message *mpa;
943 PDBG("%s ep %p\n", __func__, ep);
946 * Stop mpa timer. If it expired, then the state has
947 * changed and we bail since ep_timeout already aborted
951 if (state_read(&ep->com) != MPA_REQ_WAIT)
955 * If we get more than the supported amount of private data
956 * then we must fail this connection.
958 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
959 abort_connection(ep, skb, GFP_KERNEL);
963 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
966 * Copy the new data into our accumulation buffer.
968 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
970 ep->mpa_pkt_len += skb->len;
973 * If we don't even have the mpa message, then bail.
974 * We'll continue process when more data arrives.
976 if (ep->mpa_pkt_len < sizeof(*mpa))
978 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
979 mpa = (struct mpa_message *) ep->mpa_pkt;
982 * Validate MPA Header.
984 if (mpa->revision != mpa_rev) {
985 abort_connection(ep, skb, GFP_KERNEL);
989 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
990 abort_connection(ep, skb, GFP_KERNEL);
994 plen = ntohs(mpa->private_data_size);
997 * Fail if there's too much private data.
999 if (plen > MPA_MAX_PRIVATE_DATA) {
1000 abort_connection(ep, skb, GFP_KERNEL);
1005 * If plen does not account for pkt size
1007 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1008 abort_connection(ep, skb, GFP_KERNEL);
1011 ep->plen = (u8) plen;
1014 * If we don't have all the pdata yet, then bail.
1016 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1020 * If we get here we have accumulated the entire mpa
1021 * start reply message including private data.
1023 ep->mpa_attr.initiator = 0;
1024 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1025 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1026 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1027 ep->mpa_attr.version = mpa_rev;
1028 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1029 "xmit_marker_enabled=%d, version=%d\n", __func__,
1030 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1031 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1033 state_set(&ep->com, MPA_REQ_RCVD);
1036 connect_request_upcall(ep);
1040 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1042 struct iwch_ep *ep = ctx;
1043 struct cpl_rx_data *hdr = cplhdr(skb);
1044 unsigned int dlen = ntohs(hdr->len);
1046 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1048 skb_pull(skb, sizeof(*hdr));
1049 skb_trim(skb, dlen);
1051 ep->rcv_seq += dlen;
1052 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1054 switch (state_read(&ep->com)) {
1056 process_mpa_reply(ep, skb);
1059 process_mpa_request(ep, skb);
1064 printk(KERN_ERR MOD "%s Unexpected streaming data."
1065 " ep %p state %d tid %d\n",
1066 __func__, ep, state_read(&ep->com), ep->hwtid);
1069 * The ep will timeout and inform the ULP of the failure.
1075 /* update RX credits */
1076 update_rx_credits(ep, dlen);
1078 return CPL_RET_BUF_DONE;
1082 * Upcall from the adapter indicating data has been transmitted.
1083 * For us its just the single MPA request or reply. We can now free
1084 * the skb holding the mpa message.
1086 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1088 struct iwch_ep *ep = ctx;
1089 struct cpl_wr_ack *hdr = cplhdr(skb);
1090 unsigned int credits = ntohs(hdr->credits);
1091 unsigned long flags;
1094 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1097 PDBG("%s 0 credit ack ep %p state %u\n",
1098 __func__, ep, state_read(&ep->com));
1099 return CPL_RET_BUF_DONE;
1102 spin_lock_irqsave(&ep->com.lock, flags);
1103 BUG_ON(credits != 1);
1104 dst_confirm(ep->dst);
1106 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1107 __func__, ep, ep->com.state);
1108 if (ep->mpa_attr.initiator) {
1109 PDBG("%s initiator ep %p state %u\n",
1110 __func__, ep, ep->com.state);
1111 if (peer2peer && ep->com.state == FPDU_MODE)
1114 PDBG("%s responder ep %p state %u\n",
1115 __func__, ep, ep->com.state);
1116 if (ep->com.state == MPA_REQ_RCVD) {
1117 ep->com.rpl_done = 1;
1118 wake_up(&ep->com.waitq);
1122 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1123 __func__, ep, ep->com.state);
1124 kfree_skb(ep->mpa_skb);
1127 spin_unlock_irqrestore(&ep->com.lock, flags);
1129 iwch_post_zb_read(ep);
1130 return CPL_RET_BUF_DONE;
1133 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1135 struct iwch_ep *ep = ctx;
1136 unsigned long flags;
1139 PDBG("%s ep %p\n", __func__, ep);
1143 * We get 2 abort replies from the HW. The first one must
1144 * be ignored except for scribbling that we need one more.
1146 if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1147 return CPL_RET_BUF_DONE;
1150 spin_lock_irqsave(&ep->com.lock, flags);
1151 switch (ep->com.state) {
1153 close_complete_upcall(ep);
1154 __state_set(&ep->com, DEAD);
1158 printk(KERN_ERR "%s ep %p state %d\n",
1159 __func__, ep, ep->com.state);
1162 spin_unlock_irqrestore(&ep->com.lock, flags);
1165 release_ep_resources(ep);
1166 return CPL_RET_BUF_DONE;
1170 * Return whether a failed active open has allocated a TID
1172 static inline int act_open_has_tid(int status)
1174 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1175 status != CPL_ERR_ARP_MISS;
1178 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1180 struct iwch_ep *ep = ctx;
1181 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1183 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1184 status2errno(rpl->status));
1185 connect_reply_upcall(ep, status2errno(rpl->status));
1186 state_set(&ep->com, DEAD);
1187 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1188 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1189 cxgb3_free_atid(ep->com.tdev, ep->atid);
1190 dst_release(ep->dst);
1191 l2t_release(ep->com.tdev, ep->l2t);
1193 return CPL_RET_BUF_DONE;
1196 static int listen_start(struct iwch_listen_ep *ep)
1198 struct sk_buff *skb;
1199 struct cpl_pass_open_req *req;
1201 PDBG("%s ep %p\n", __func__, ep);
1202 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1204 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1208 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1209 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1210 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1211 req->local_port = ep->com.local_addr.sin_port;
1212 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1215 req->peer_netmask = 0;
1216 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1217 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1218 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1221 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1224 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1226 struct iwch_listen_ep *ep = ctx;
1227 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1229 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1230 rpl->status, status2errno(rpl->status));
1231 ep->com.rpl_err = status2errno(rpl->status);
1232 ep->com.rpl_done = 1;
1233 wake_up(&ep->com.waitq);
1235 return CPL_RET_BUF_DONE;
1238 static int listen_stop(struct iwch_listen_ep *ep)
1240 struct sk_buff *skb;
1241 struct cpl_close_listserv_req *req;
1243 PDBG("%s ep %p\n", __func__, ep);
1244 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1246 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1249 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1250 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1252 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1254 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1257 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1260 struct iwch_listen_ep *ep = ctx;
1261 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1263 PDBG("%s ep %p\n", __func__, ep);
1264 ep->com.rpl_err = status2errno(rpl->status);
1265 ep->com.rpl_done = 1;
1266 wake_up(&ep->com.waitq);
1267 return CPL_RET_BUF_DONE;
1270 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1272 struct cpl_pass_accept_rpl *rpl;
1273 unsigned int mtu_idx;
1274 u32 opt0h, opt0l, opt2;
1277 PDBG("%s ep %p\n", __func__, ep);
1278 BUG_ON(skb_cloned(skb));
1279 skb_trim(skb, sizeof(*rpl));
1281 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1282 wscale = compute_wscale(rcv_win);
1283 opt0h = V_NAGLE(0) |
1287 V_WND_SCALE(wscale) |
1288 V_MSS_IDX(mtu_idx) |
1289 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1290 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1291 opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1292 V_CONG_CONTROL_FLAVOR(cong_flavor);
1295 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1296 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1297 rpl->peer_ip = peer_ip;
1298 rpl->opt0h = htonl(opt0h);
1299 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1300 rpl->opt2 = htonl(opt2);
1301 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1302 skb->priority = CPL_PRIORITY_SETUP;
1303 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1308 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1309 struct sk_buff *skb)
1311 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1313 BUG_ON(skb_cloned(skb));
1314 skb_trim(skb, sizeof(struct cpl_tid_release));
1317 if (tdev->type != T3A)
1318 release_tid(tdev, hwtid, skb);
1320 struct cpl_pass_accept_rpl *rpl;
1323 skb->priority = CPL_PRIORITY_SETUP;
1324 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1325 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1327 rpl->peer_ip = peer_ip;
1328 rpl->opt0h = htonl(F_TCAM_BYPASS);
1329 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1331 rpl->rsvd = rpl->opt2;
1332 iwch_cxgb3_ofld_send(tdev, skb);
1336 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1338 struct iwch_ep *child_ep, *parent_ep = ctx;
1339 struct cpl_pass_accept_req *req = cplhdr(skb);
1340 unsigned int hwtid = GET_TID(req);
1341 struct neighbour *neigh;
1342 struct dst_entry *dst;
1343 struct l2t_entry *l2t;
1347 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1349 if (state_read(&parent_ep->com) != LISTEN) {
1350 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1356 * Find the netdev for this connection request.
1358 tim.mac_addr = req->dst_mac;
1359 tim.vlan_tag = ntohs(req->vlan_tag);
1360 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1361 printk(KERN_ERR "%s bad dst mac %pM\n",
1362 __func__, req->dst_mac);
1366 /* Find output route */
1367 rt = find_route(tdev,
1371 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1373 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1378 neigh = dst_get_neighbour(dst);
1379 l2t = t3_l2t_get(tdev, neigh, neigh->dev);
1381 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1386 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1388 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1390 l2t_release(tdev, l2t);
1394 state_set(&child_ep->com, CONNECTING);
1395 child_ep->com.tdev = tdev;
1396 child_ep->com.cm_id = NULL;
1397 child_ep->com.local_addr.sin_family = PF_INET;
1398 child_ep->com.local_addr.sin_port = req->local_port;
1399 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1400 child_ep->com.remote_addr.sin_family = PF_INET;
1401 child_ep->com.remote_addr.sin_port = req->peer_port;
1402 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1403 get_ep(&parent_ep->com);
1404 child_ep->parent_ep = parent_ep;
1405 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1406 child_ep->l2t = l2t;
1407 child_ep->dst = dst;
1408 child_ep->hwtid = hwtid;
1409 init_timer(&child_ep->timer);
1410 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1411 accept_cr(child_ep, req->peer_ip, skb);
1414 reject_cr(tdev, hwtid, req->peer_ip, skb);
1416 return CPL_RET_BUF_DONE;
1419 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1421 struct iwch_ep *ep = ctx;
1422 struct cpl_pass_establish *req = cplhdr(skb);
1424 PDBG("%s ep %p\n", __func__, ep);
1425 ep->snd_seq = ntohl(req->snd_isn);
1426 ep->rcv_seq = ntohl(req->rcv_isn);
1428 set_emss(ep, ntohs(req->tcp_opt));
1430 dst_confirm(ep->dst);
1431 state_set(&ep->com, MPA_REQ_WAIT);
1434 return CPL_RET_BUF_DONE;
1437 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1439 struct iwch_ep *ep = ctx;
1440 struct iwch_qp_attributes attrs;
1441 unsigned long flags;
1445 PDBG("%s ep %p\n", __func__, ep);
1446 dst_confirm(ep->dst);
1448 spin_lock_irqsave(&ep->com.lock, flags);
1449 switch (ep->com.state) {
1451 __state_set(&ep->com, CLOSING);
1454 __state_set(&ep->com, CLOSING);
1455 connect_reply_upcall(ep, -ECONNRESET);
1460 * We're gonna mark this puppy DEAD, but keep
1461 * the reference on it until the ULP accepts or
1462 * rejects the CR. Also wake up anyone waiting
1463 * in rdma connection migration (see iwch_accept_cr()).
1465 __state_set(&ep->com, CLOSING);
1466 ep->com.rpl_done = 1;
1467 ep->com.rpl_err = -ECONNRESET;
1468 PDBG("waking up ep %p\n", ep);
1469 wake_up(&ep->com.waitq);
1472 __state_set(&ep->com, CLOSING);
1473 ep->com.rpl_done = 1;
1474 ep->com.rpl_err = -ECONNRESET;
1475 PDBG("waking up ep %p\n", ep);
1476 wake_up(&ep->com.waitq);
1480 __state_set(&ep->com, CLOSING);
1481 attrs.next_state = IWCH_QP_STATE_CLOSING;
1482 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1483 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1484 peer_close_upcall(ep);
1490 __state_set(&ep->com, MORIBUND);
1495 if (ep->com.cm_id && ep->com.qp) {
1496 attrs.next_state = IWCH_QP_STATE_IDLE;
1497 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1498 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1500 close_complete_upcall(ep);
1501 __state_set(&ep->com, DEAD);
1511 spin_unlock_irqrestore(&ep->com.lock, flags);
1513 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1515 release_ep_resources(ep);
1516 return CPL_RET_BUF_DONE;
1520 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1522 static int is_neg_adv_abort(unsigned int status)
1524 return status == CPL_ERR_RTX_NEG_ADVICE ||
1525 status == CPL_ERR_PERSIST_NEG_ADVICE;
1528 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1530 struct cpl_abort_req_rss *req = cplhdr(skb);
1531 struct iwch_ep *ep = ctx;
1532 struct cpl_abort_rpl *rpl;
1533 struct sk_buff *rpl_skb;
1534 struct iwch_qp_attributes attrs;
1537 unsigned long flags;
1539 if (is_neg_adv_abort(req->status)) {
1540 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1542 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1543 return CPL_RET_BUF_DONE;
1547 * We get 2 peer aborts from the HW. The first one must
1548 * be ignored except for scribbling that we need one more.
1550 if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
1551 return CPL_RET_BUF_DONE;
1554 spin_lock_irqsave(&ep->com.lock, flags);
1555 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1556 switch (ep->com.state) {
1564 connect_reply_upcall(ep, -ECONNRESET);
1567 ep->com.rpl_done = 1;
1568 ep->com.rpl_err = -ECONNRESET;
1569 PDBG("waking up ep %p\n", ep);
1570 wake_up(&ep->com.waitq);
1575 * We're gonna mark this puppy DEAD, but keep
1576 * the reference on it until the ULP accepts or
1577 * rejects the CR. Also wake up anyone waiting
1578 * in rdma connection migration (see iwch_accept_cr()).
1580 ep->com.rpl_done = 1;
1581 ep->com.rpl_err = -ECONNRESET;
1582 PDBG("waking up ep %p\n", ep);
1583 wake_up(&ep->com.waitq);
1590 if (ep->com.cm_id && ep->com.qp) {
1591 attrs.next_state = IWCH_QP_STATE_ERROR;
1592 ret = iwch_modify_qp(ep->com.qp->rhp,
1593 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1597 "%s - qp <- error failed!\n",
1600 peer_abort_upcall(ep);
1605 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1606 spin_unlock_irqrestore(&ep->com.lock, flags);
1607 return CPL_RET_BUF_DONE;
1612 dst_confirm(ep->dst);
1613 if (ep->com.state != ABORTING) {
1614 __state_set(&ep->com, DEAD);
1617 spin_unlock_irqrestore(&ep->com.lock, flags);
1619 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1621 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1626 rpl_skb->priority = CPL_PRIORITY_DATA;
1627 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1628 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1629 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1630 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1631 rpl->cmd = CPL_ABORT_NO_RST;
1632 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1635 release_ep_resources(ep);
1636 return CPL_RET_BUF_DONE;
1639 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1641 struct iwch_ep *ep = ctx;
1642 struct iwch_qp_attributes attrs;
1643 unsigned long flags;
1646 PDBG("%s ep %p\n", __func__, ep);
1649 /* The cm_id may be null if we failed to connect */
1650 spin_lock_irqsave(&ep->com.lock, flags);
1651 switch (ep->com.state) {
1653 __state_set(&ep->com, MORIBUND);
1657 if ((ep->com.cm_id) && (ep->com.qp)) {
1658 attrs.next_state = IWCH_QP_STATE_IDLE;
1659 iwch_modify_qp(ep->com.qp->rhp,
1661 IWCH_QP_ATTR_NEXT_STATE,
1664 close_complete_upcall(ep);
1665 __state_set(&ep->com, DEAD);
1675 spin_unlock_irqrestore(&ep->com.lock, flags);
1677 release_ep_resources(ep);
1678 return CPL_RET_BUF_DONE;
1682 * T3A does 3 things when a TERM is received:
1683 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1684 * 2) generate an async event on the QP with the TERMINATE opcode
1685 * 3) post a TERMINATE opcde cqe into the associated CQ.
1687 * For (1), we save the message in the qp for later consumer consumption.
1688 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1689 * For (3), we toss the CQE in cxio_poll_cq().
1691 * terminate() handles case (1)...
1693 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1695 struct iwch_ep *ep = ctx;
1697 if (state_read(&ep->com) != FPDU_MODE)
1698 return CPL_RET_BUF_DONE;
1700 PDBG("%s ep %p\n", __func__, ep);
1701 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1702 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1703 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1705 ep->com.qp->attr.terminate_msg_len = skb->len;
1706 ep->com.qp->attr.is_terminate_local = 0;
1707 return CPL_RET_BUF_DONE;
1710 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1712 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1713 struct iwch_ep *ep = ctx;
1715 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
1718 struct iwch_qp_attributes attrs;
1720 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1721 __func__, ep->hwtid);
1723 attrs.next_state = IWCH_QP_STATE_ERROR;
1724 iwch_modify_qp(ep->com.qp->rhp,
1725 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1727 abort_connection(ep, NULL, GFP_KERNEL);
1729 return CPL_RET_BUF_DONE;
1732 static void ep_timeout(unsigned long arg)
1734 struct iwch_ep *ep = (struct iwch_ep *)arg;
1735 struct iwch_qp_attributes attrs;
1736 unsigned long flags;
1739 spin_lock_irqsave(&ep->com.lock, flags);
1740 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
1742 switch (ep->com.state) {
1744 __state_set(&ep->com, ABORTING);
1745 connect_reply_upcall(ep, -ETIMEDOUT);
1748 __state_set(&ep->com, ABORTING);
1752 if (ep->com.cm_id && ep->com.qp) {
1753 attrs.next_state = IWCH_QP_STATE_ERROR;
1754 iwch_modify_qp(ep->com.qp->rhp,
1755 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1758 __state_set(&ep->com, ABORTING);
1761 printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1762 __func__, ep, ep->com.state);
1766 spin_unlock_irqrestore(&ep->com.lock, flags);
1768 abort_connection(ep, NULL, GFP_ATOMIC);
1772 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1775 struct iwch_ep *ep = to_ep(cm_id);
1776 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1778 if (state_read(&ep->com) == DEAD) {
1782 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1784 abort_connection(ep, NULL, GFP_KERNEL);
1786 err = send_mpa_reject(ep, pdata, pdata_len);
1787 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1793 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1796 struct iwch_qp_attributes attrs;
1797 enum iwch_qp_attr_mask mask;
1798 struct iwch_ep *ep = to_ep(cm_id);
1799 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1800 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1802 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1803 if (state_read(&ep->com) == DEAD) {
1808 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1811 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1812 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1813 abort_connection(ep, NULL, GFP_KERNEL);
1818 cm_id->add_ref(cm_id);
1819 ep->com.cm_id = cm_id;
1822 ep->ird = conn_param->ird;
1823 ep->ord = conn_param->ord;
1825 if (peer2peer && ep->ird == 0)
1828 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1830 /* bind QP to EP and move to RTS */
1831 attrs.mpa_attr = ep->mpa_attr;
1832 attrs.max_ird = ep->ird;
1833 attrs.max_ord = ep->ord;
1834 attrs.llp_stream_handle = ep;
1835 attrs.next_state = IWCH_QP_STATE_RTS;
1837 /* bind QP and TID with INIT_WR */
1838 mask = IWCH_QP_ATTR_NEXT_STATE |
1839 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1840 IWCH_QP_ATTR_MPA_ATTR |
1841 IWCH_QP_ATTR_MAX_IRD |
1842 IWCH_QP_ATTR_MAX_ORD;
1844 err = iwch_modify_qp(ep->com.qp->rhp,
1845 ep->com.qp, mask, &attrs, 1);
1849 /* if needed, wait for wr_ack */
1850 if (iwch_rqes_posted(qp)) {
1851 wait_event(ep->com.waitq, ep->com.rpl_done);
1852 err = ep->com.rpl_err;
1857 err = send_mpa_reply(ep, conn_param->private_data,
1858 conn_param->private_data_len);
1863 state_set(&ep->com, FPDU_MODE);
1864 established_upcall(ep);
1868 ep->com.cm_id = NULL;
1870 cm_id->rem_ref(cm_id);
1876 static int is_loopback_dst(struct iw_cm_id *cm_id)
1878 struct net_device *dev;
1880 dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1887 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1889 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1890 struct neighbour *neigh;
1895 if (is_loopback_dst(cm_id)) {
1900 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1902 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1906 init_timer(&ep->timer);
1907 ep->plen = conn_param->private_data_len;
1909 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1910 conn_param->private_data, ep->plen);
1911 ep->ird = conn_param->ird;
1912 ep->ord = conn_param->ord;
1914 if (peer2peer && ep->ord == 0)
1917 ep->com.tdev = h->rdev.t3cdev_p;
1919 cm_id->add_ref(cm_id);
1920 ep->com.cm_id = cm_id;
1921 ep->com.qp = get_qhp(h, conn_param->qpn);
1922 BUG_ON(!ep->com.qp);
1923 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1927 * Allocate an active TID to initiate a TCP connection.
1929 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1930 if (ep->atid == -1) {
1931 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1937 rt = find_route(h->rdev.t3cdev_p,
1938 cm_id->local_addr.sin_addr.s_addr,
1939 cm_id->remote_addr.sin_addr.s_addr,
1940 cm_id->local_addr.sin_port,
1941 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1943 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1944 err = -EHOSTUNREACH;
1949 neigh = dst_get_neighbour(ep->dst);
1951 /* get a l2t entry */
1952 ep->l2t = t3_l2t_get(ep->com.tdev, neigh, neigh->dev);
1954 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1959 state_set(&ep->com, CONNECTING);
1960 ep->tos = IPTOS_LOWDELAY;
1961 ep->com.local_addr = cm_id->local_addr;
1962 ep->com.remote_addr = cm_id->remote_addr;
1964 /* send connect request to rnic */
1965 err = send_connect(ep);
1969 l2t_release(h->rdev.t3cdev_p, ep->l2t);
1971 dst_release(ep->dst);
1973 cxgb3_free_atid(ep->com.tdev, ep->atid);
1975 cm_id->rem_ref(cm_id);
1981 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1984 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1985 struct iwch_listen_ep *ep;
1990 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1992 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1996 PDBG("%s ep %p\n", __func__, ep);
1997 ep->com.tdev = h->rdev.t3cdev_p;
1998 cm_id->add_ref(cm_id);
1999 ep->com.cm_id = cm_id;
2000 ep->backlog = backlog;
2001 ep->com.local_addr = cm_id->local_addr;
2004 * Allocate a server TID.
2006 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
2007 if (ep->stid == -1) {
2008 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2013 state_set(&ep->com, LISTEN);
2014 err = listen_start(ep);
2018 /* wait for pass_open_rpl */
2019 wait_event(ep->com.waitq, ep->com.rpl_done);
2020 err = ep->com.rpl_err;
2022 cm_id->provider_data = ep;
2026 cxgb3_free_stid(ep->com.tdev, ep->stid);
2028 cm_id->rem_ref(cm_id);
2035 int iwch_destroy_listen(struct iw_cm_id *cm_id)
2038 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2040 PDBG("%s ep %p\n", __func__, ep);
2043 state_set(&ep->com, DEAD);
2044 ep->com.rpl_done = 0;
2045 ep->com.rpl_err = 0;
2046 err = listen_stop(ep);
2049 wait_event(ep->com.waitq, ep->com.rpl_done);
2050 cxgb3_free_stid(ep->com.tdev, ep->stid);
2052 err = ep->com.rpl_err;
2053 cm_id->rem_ref(cm_id);
2058 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2061 unsigned long flags;
2064 struct t3cdev *tdev;
2065 struct cxio_rdev *rdev;
2067 spin_lock_irqsave(&ep->com.lock, flags);
2069 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2070 states[ep->com.state], abrupt);
2072 tdev = (struct t3cdev *)ep->com.tdev;
2073 rdev = (struct cxio_rdev *)tdev->ulp;
2074 if (cxio_fatal_error(rdev)) {
2076 close_complete_upcall(ep);
2077 ep->com.state = DEAD;
2079 switch (ep->com.state) {
2087 ep->com.state = ABORTING;
2089 ep->com.state = CLOSING;
2092 set_bit(CLOSE_SENT, &ep->com.flags);
2095 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2099 ep->com.state = ABORTING;
2101 ep->com.state = MORIBUND;
2107 PDBG("%s ignoring disconnect ep %p state %u\n",
2108 __func__, ep, ep->com.state);
2115 spin_unlock_irqrestore(&ep->com.lock, flags);
2118 ret = send_abort(ep, NULL, gfp);
2120 ret = send_halfclose(ep, gfp);
2125 release_ep_resources(ep);
2129 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2130 struct l2t_entry *l2t)
2132 struct iwch_ep *ep = ctx;
2137 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2140 l2t_release(ep->com.tdev, ep->l2t);
2148 * All the CM events are handled on a work queue to have a safe context.
2149 * These are the real handlers that are called from the work queue.
2151 static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2152 [CPL_ACT_ESTABLISH] = act_establish,
2153 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2154 [CPL_RX_DATA] = rx_data,
2155 [CPL_TX_DMA_ACK] = tx_ack,
2156 [CPL_ABORT_RPL_RSS] = abort_rpl,
2157 [CPL_ABORT_RPL] = abort_rpl,
2158 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2159 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2160 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2161 [CPL_PASS_ESTABLISH] = pass_establish,
2162 [CPL_PEER_CLOSE] = peer_close,
2163 [CPL_ABORT_REQ_RSS] = peer_abort,
2164 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2165 [CPL_RDMA_TERMINATE] = terminate,
2166 [CPL_RDMA_EC_STATUS] = ec_status,
2169 static void process_work(struct work_struct *work)
2171 struct sk_buff *skb = NULL;
2173 struct t3cdev *tdev;
2176 while ((skb = skb_dequeue(&rxq))) {
2177 ep = *((void **) (skb->cb));
2178 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2179 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2180 if (ret & CPL_RET_BUF_DONE)
2184 * ep was referenced in sched(), and is freed here.
2186 put_ep((struct iwch_ep_common *)ep);
2190 static DECLARE_WORK(skb_work, process_work);
2192 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2194 struct iwch_ep_common *epc = ctx;
2199 * Save ctx and tdev in the skb->cb area.
2201 *((void **) skb->cb) = ctx;
2202 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2205 * Queue the skb and schedule the worker thread.
2207 skb_queue_tail(&rxq, skb);
2208 queue_work(workq, &skb_work);
2212 static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2214 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2216 if (rpl->status != CPL_ERR_NONE) {
2217 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2218 "for tid %u\n", rpl->status, GET_TID(rpl));
2220 return CPL_RET_BUF_DONE;
2224 * All upcalls from the T3 Core go to sched() to schedule the
2225 * processing on a work queue.
2227 cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2228 [CPL_ACT_ESTABLISH] = sched,
2229 [CPL_ACT_OPEN_RPL] = sched,
2230 [CPL_RX_DATA] = sched,
2231 [CPL_TX_DMA_ACK] = sched,
2232 [CPL_ABORT_RPL_RSS] = sched,
2233 [CPL_ABORT_RPL] = sched,
2234 [CPL_PASS_OPEN_RPL] = sched,
2235 [CPL_CLOSE_LISTSRV_RPL] = sched,
2236 [CPL_PASS_ACCEPT_REQ] = sched,
2237 [CPL_PASS_ESTABLISH] = sched,
2238 [CPL_PEER_CLOSE] = sched,
2239 [CPL_CLOSE_CON_RPL] = sched,
2240 [CPL_ABORT_REQ_RSS] = sched,
2241 [CPL_RDMA_TERMINATE] = sched,
2242 [CPL_RDMA_EC_STATUS] = sched,
2243 [CPL_SET_TCB_RPL] = set_tcb_rpl,
2246 int __init iwch_cm_init(void)
2248 skb_queue_head_init(&rxq);
2250 workq = create_singlethread_workqueue("iw_cxgb3");
2257 void __exit iwch_cm_term(void)
2259 flush_workqueue(workq);
2260 destroy_workqueue(workq);