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>
38 #include <linux/inetdevice.h>
40 #include <net/neighbour.h>
41 #include <net/netevent.h>
42 #include <net/route.h>
45 #include "cxgb3_offload.h"
47 #include "iwch_provider.h"
50 static char *states[] = {
67 module_param(peer2peer, int, 0644);
68 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
70 static int ep_timeout_secs = 60;
71 module_param(ep_timeout_secs, int, 0644);
72 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
73 "in seconds (default=60)");
75 static int mpa_rev = 1;
76 module_param(mpa_rev, int, 0644);
77 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
78 "1 is spec compliant. (default=1)");
80 static int markers_enabled = 0;
81 module_param(markers_enabled, int, 0644);
82 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
84 static int crc_enabled = 1;
85 module_param(crc_enabled, int, 0644);
86 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
88 static int rcv_win = 256 * 1024;
89 module_param(rcv_win, int, 0644);
90 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
92 static int snd_win = 32 * 1024;
93 module_param(snd_win, int, 0644);
94 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
96 static unsigned int nocong = 0;
97 module_param(nocong, uint, 0644);
98 MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
100 static unsigned int cong_flavor = 1;
101 module_param(cong_flavor, uint, 0644);
102 MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
104 static void process_work(struct work_struct *work);
105 static struct workqueue_struct *workq;
106 static DECLARE_WORK(skb_work, process_work);
108 static struct sk_buff_head rxq;
109 static cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS];
111 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
112 static void ep_timeout(unsigned long arg);
113 static void connect_reply_upcall(struct iwch_ep *ep, int status);
115 static void start_ep_timer(struct iwch_ep *ep)
117 PDBG("%s ep %p\n", __func__, ep);
118 if (timer_pending(&ep->timer)) {
119 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
120 del_timer_sync(&ep->timer);
123 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
124 ep->timer.data = (unsigned long)ep;
125 ep->timer.function = ep_timeout;
126 add_timer(&ep->timer);
129 static void stop_ep_timer(struct iwch_ep *ep)
131 PDBG("%s ep %p\n", __func__, ep);
132 if (!timer_pending(&ep->timer)) {
133 printk(KERN_ERR "%s timer stopped when its not running! ep %p state %u\n",
134 __func__, ep, ep->com.state);
138 del_timer_sync(&ep->timer);
142 int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
145 struct cxio_rdev *rdev;
147 rdev = (struct cxio_rdev *)tdev->ulp;
148 if (cxio_fatal_error(rdev)) {
152 error = l2t_send(tdev, skb, l2e);
158 int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
161 struct cxio_rdev *rdev;
163 rdev = (struct cxio_rdev *)tdev->ulp;
164 if (cxio_fatal_error(rdev)) {
168 error = cxgb3_ofld_send(tdev, skb);
174 static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
176 struct cpl_tid_release *req;
178 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
181 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
182 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
183 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
184 skb->priority = CPL_PRIORITY_SETUP;
185 iwch_cxgb3_ofld_send(tdev, skb);
189 int iwch_quiesce_tid(struct iwch_ep *ep)
191 struct cpl_set_tcb_field *req;
192 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
196 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
197 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
198 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
199 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
202 req->word = htons(W_TCB_RX_QUIESCE);
203 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
204 req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
206 skb->priority = CPL_PRIORITY_DATA;
207 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
210 int iwch_resume_tid(struct iwch_ep *ep)
212 struct cpl_set_tcb_field *req;
213 struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
217 req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
218 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
219 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
220 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
223 req->word = htons(W_TCB_RX_QUIESCE);
224 req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
227 skb->priority = CPL_PRIORITY_DATA;
228 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
231 static void set_emss(struct iwch_ep *ep, u16 opt)
233 PDBG("%s ep %p opt %u\n", __func__, ep, opt);
234 ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
235 if (G_TCPOPT_TSTAMP(opt))
239 PDBG("emss=%d\n", ep->emss);
242 static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
245 enum iwch_ep_state state;
247 spin_lock_irqsave(&epc->lock, flags);
249 spin_unlock_irqrestore(&epc->lock, flags);
253 static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
258 static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
262 spin_lock_irqsave(&epc->lock, flags);
263 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
264 __state_set(epc, new);
265 spin_unlock_irqrestore(&epc->lock, flags);
269 static void *alloc_ep(int size, gfp_t gfp)
271 struct iwch_ep_common *epc;
273 epc = kzalloc(size, gfp);
275 kref_init(&epc->kref);
276 spin_lock_init(&epc->lock);
277 init_waitqueue_head(&epc->waitq);
279 PDBG("%s alloc ep %p\n", __func__, epc);
283 void __free_ep(struct kref *kref)
286 ep = container_of(container_of(kref, struct iwch_ep_common, kref),
287 struct iwch_ep, com);
288 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
289 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
290 cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
291 dst_release(ep->dst);
292 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
297 static void release_ep_resources(struct iwch_ep *ep)
299 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
300 set_bit(RELEASE_RESOURCES, &ep->com.flags);
304 static void process_work(struct work_struct *work)
306 struct sk_buff *skb = NULL;
311 while ((skb = skb_dequeue(&rxq))) {
312 ep = *((void **) (skb->cb));
313 tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
314 ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
315 if (ret & CPL_RET_BUF_DONE)
319 * ep was referenced in sched(), and is freed here.
321 put_ep((struct iwch_ep_common *)ep);
325 static int status2errno(int status)
330 case CPL_ERR_CONN_RESET:
332 case CPL_ERR_ARP_MISS:
333 return -EHOSTUNREACH;
334 case CPL_ERR_CONN_TIMEDOUT:
336 case CPL_ERR_TCAM_FULL:
338 case CPL_ERR_CONN_EXIST:
346 * Try and reuse skbs already allocated...
348 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
350 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
354 skb = alloc_skb(len, gfp);
359 static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
360 __be32 peer_ip, __be16 local_port,
361 __be16 peer_port, u8 tos)
372 .proto = IPPROTO_TCP,
380 if (ip_route_output_flow(&init_net, &rt, &fl, NULL, 0))
385 static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
389 while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
394 static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
396 PDBG("%s t3cdev %p\n", __func__, dev);
401 * Handle an ARP failure for an active open.
403 static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
405 printk(KERN_ERR MOD "ARP failure duing connect\n");
410 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
413 static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
415 struct cpl_abort_req *req = cplhdr(skb);
417 PDBG("%s t3cdev %p\n", __func__, dev);
418 req->cmd = CPL_ABORT_NO_RST;
419 iwch_cxgb3_ofld_send(dev, skb);
422 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
424 struct cpl_close_con_req *req;
427 PDBG("%s ep %p\n", __func__, ep);
428 skb = get_skb(NULL, sizeof(*req), gfp);
430 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
433 skb->priority = CPL_PRIORITY_DATA;
434 set_arp_failure_handler(skb, arp_failure_discard);
435 req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
436 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
437 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
438 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
439 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
442 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
444 struct cpl_abort_req *req;
446 PDBG("%s ep %p\n", __func__, ep);
447 skb = get_skb(skb, sizeof(*req), gfp);
449 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
453 skb->priority = CPL_PRIORITY_DATA;
454 set_arp_failure_handler(skb, abort_arp_failure);
455 req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
456 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
457 req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
458 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
459 req->cmd = CPL_ABORT_SEND_RST;
460 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
463 static int send_connect(struct iwch_ep *ep)
465 struct cpl_act_open_req *req;
467 u32 opt0h, opt0l, opt2;
468 unsigned int mtu_idx;
471 PDBG("%s ep %p\n", __func__, ep);
473 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
475 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
479 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
480 wscale = compute_wscale(rcv_win);
485 V_WND_SCALE(wscale) |
487 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
488 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
489 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
490 skb->priority = CPL_PRIORITY_SETUP;
491 set_arp_failure_handler(skb, act_open_req_arp_failure);
493 req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
494 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
495 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
496 req->local_port = ep->com.local_addr.sin_port;
497 req->peer_port = ep->com.remote_addr.sin_port;
498 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
499 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
500 req->opt0h = htonl(opt0h);
501 req->opt0l = htonl(opt0l);
503 req->opt2 = htonl(opt2);
504 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
507 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
510 struct tx_data_wr *req;
511 struct mpa_message *mpa;
514 PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
516 BUG_ON(skb_cloned(skb));
518 mpalen = sizeof(*mpa) + ep->plen;
519 if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
521 skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
523 connect_reply_upcall(ep, -ENOMEM);
528 skb_reserve(skb, sizeof(*req));
529 skb_put(skb, mpalen);
530 skb->priority = CPL_PRIORITY_DATA;
531 mpa = (struct mpa_message *) skb->data;
532 memset(mpa, 0, sizeof(*mpa));
533 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
534 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
535 (markers_enabled ? MPA_MARKERS : 0);
536 mpa->private_data_size = htons(ep->plen);
537 mpa->revision = mpa_rev;
540 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
543 * Reference the mpa skb. This ensures the data area
544 * will remain in memory until the hw acks the tx.
545 * Function tx_ack() will deref it.
548 set_arp_failure_handler(skb, arp_failure_discard);
549 skb_reset_transport_header(skb);
551 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
552 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
553 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
554 req->len = htonl(len);
555 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
556 V_TX_SNDBUF(snd_win>>15));
557 req->flags = htonl(F_TX_INIT);
558 req->sndseq = htonl(ep->snd_seq);
561 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
563 state_set(&ep->com, MPA_REQ_SENT);
567 static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
570 struct tx_data_wr *req;
571 struct mpa_message *mpa;
574 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
576 mpalen = sizeof(*mpa) + plen;
578 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
580 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
583 skb_reserve(skb, sizeof(*req));
584 mpa = (struct mpa_message *) skb_put(skb, mpalen);
585 memset(mpa, 0, sizeof(*mpa));
586 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
587 mpa->flags = MPA_REJECT;
588 mpa->revision = mpa_rev;
589 mpa->private_data_size = htons(plen);
591 memcpy(mpa->private_data, pdata, plen);
594 * Reference the mpa skb again. This ensures the data area
595 * will remain in memory until the hw acks the tx.
596 * Function tx_ack() will deref it.
599 skb->priority = CPL_PRIORITY_DATA;
600 set_arp_failure_handler(skb, arp_failure_discard);
601 skb_reset_transport_header(skb);
602 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
603 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
604 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
605 req->len = htonl(mpalen);
606 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
607 V_TX_SNDBUF(snd_win>>15));
608 req->flags = htonl(F_TX_INIT);
609 req->sndseq = htonl(ep->snd_seq);
612 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
615 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
618 struct tx_data_wr *req;
619 struct mpa_message *mpa;
623 PDBG("%s ep %p plen %d\n", __func__, ep, plen);
625 mpalen = sizeof(*mpa) + plen;
627 skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
629 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
632 skb->priority = CPL_PRIORITY_DATA;
633 skb_reserve(skb, sizeof(*req));
634 mpa = (struct mpa_message *) skb_put(skb, mpalen);
635 memset(mpa, 0, sizeof(*mpa));
636 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
637 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
638 (markers_enabled ? MPA_MARKERS : 0);
639 mpa->revision = mpa_rev;
640 mpa->private_data_size = htons(plen);
642 memcpy(mpa->private_data, pdata, plen);
645 * Reference the mpa skb. This ensures the data area
646 * will remain in memory until the hw acks the tx.
647 * Function tx_ack() will deref it.
650 set_arp_failure_handler(skb, arp_failure_discard);
651 skb_reset_transport_header(skb);
653 req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
654 req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
655 req->wr_lo = htonl(V_WR_TID(ep->hwtid));
656 req->len = htonl(len);
657 req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
658 V_TX_SNDBUF(snd_win>>15));
659 req->flags = htonl(F_TX_INIT);
660 req->sndseq = htonl(ep->snd_seq);
662 state_set(&ep->com, MPA_REP_SENT);
663 return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
666 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
668 struct iwch_ep *ep = ctx;
669 struct cpl_act_establish *req = cplhdr(skb);
670 unsigned int tid = GET_TID(req);
672 PDBG("%s ep %p tid %d\n", __func__, ep, tid);
674 dst_confirm(ep->dst);
676 /* setup the hwtid for this connection */
678 cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
680 ep->snd_seq = ntohl(req->snd_isn);
681 ep->rcv_seq = ntohl(req->rcv_isn);
683 set_emss(ep, ntohs(req->tcp_opt));
685 /* dealloc the atid */
686 cxgb3_free_atid(ep->com.tdev, ep->atid);
688 /* start MPA negotiation */
689 send_mpa_req(ep, skb);
694 static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
696 PDBG("%s ep %p\n", __FILE__, ep);
697 state_set(&ep->com, ABORTING);
698 send_abort(ep, skb, gfp);
701 static void close_complete_upcall(struct iwch_ep *ep)
703 struct iw_cm_event event;
705 PDBG("%s ep %p\n", __func__, ep);
706 memset(&event, 0, sizeof(event));
707 event.event = IW_CM_EVENT_CLOSE;
709 PDBG("close complete delivered ep %p cm_id %p tid %d\n",
710 ep, ep->com.cm_id, ep->hwtid);
711 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
712 ep->com.cm_id->rem_ref(ep->com.cm_id);
713 ep->com.cm_id = NULL;
718 static void peer_close_upcall(struct iwch_ep *ep)
720 struct iw_cm_event event;
722 PDBG("%s ep %p\n", __func__, ep);
723 memset(&event, 0, sizeof(event));
724 event.event = IW_CM_EVENT_DISCONNECT;
726 PDBG("peer close delivered ep %p cm_id %p tid %d\n",
727 ep, ep->com.cm_id, ep->hwtid);
728 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
732 static void peer_abort_upcall(struct iwch_ep *ep)
734 struct iw_cm_event event;
736 PDBG("%s ep %p\n", __func__, ep);
737 memset(&event, 0, sizeof(event));
738 event.event = IW_CM_EVENT_CLOSE;
739 event.status = -ECONNRESET;
741 PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
742 ep->com.cm_id, ep->hwtid);
743 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
744 ep->com.cm_id->rem_ref(ep->com.cm_id);
745 ep->com.cm_id = NULL;
750 static void connect_reply_upcall(struct iwch_ep *ep, int status)
752 struct iw_cm_event event;
754 PDBG("%s ep %p status %d\n", __func__, ep, status);
755 memset(&event, 0, sizeof(event));
756 event.event = IW_CM_EVENT_CONNECT_REPLY;
757 event.status = status;
758 event.local_addr = ep->com.local_addr;
759 event.remote_addr = ep->com.remote_addr;
761 if ((status == 0) || (status == -ECONNREFUSED)) {
762 event.private_data_len = ep->plen;
763 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
766 PDBG("%s ep %p tid %d status %d\n", __func__, ep,
768 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
771 ep->com.cm_id->rem_ref(ep->com.cm_id);
772 ep->com.cm_id = NULL;
777 static void connect_request_upcall(struct iwch_ep *ep)
779 struct iw_cm_event event;
781 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
782 memset(&event, 0, sizeof(event));
783 event.event = IW_CM_EVENT_CONNECT_REQUEST;
784 event.local_addr = ep->com.local_addr;
785 event.remote_addr = ep->com.remote_addr;
786 event.private_data_len = ep->plen;
787 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
788 event.provider_data = ep;
789 if (state_read(&ep->parent_ep->com) != DEAD) {
791 ep->parent_ep->com.cm_id->event_handler(
792 ep->parent_ep->com.cm_id,
795 put_ep(&ep->parent_ep->com);
796 ep->parent_ep = NULL;
799 static void established_upcall(struct iwch_ep *ep)
801 struct iw_cm_event event;
803 PDBG("%s ep %p\n", __func__, ep);
804 memset(&event, 0, sizeof(event));
805 event.event = IW_CM_EVENT_ESTABLISHED;
807 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
808 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
812 static int update_rx_credits(struct iwch_ep *ep, u32 credits)
814 struct cpl_rx_data_ack *req;
817 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
818 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
820 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
824 req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
825 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
826 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
827 req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
828 skb->priority = CPL_PRIORITY_ACK;
829 iwch_cxgb3_ofld_send(ep->com.tdev, skb);
833 static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
835 struct mpa_message *mpa;
837 struct iwch_qp_attributes attrs;
838 enum iwch_qp_attr_mask mask;
841 PDBG("%s ep %p\n", __func__, ep);
844 * Stop mpa timer. If it expired, then the state has
845 * changed and we bail since ep_timeout already aborted
849 if (state_read(&ep->com) != MPA_REQ_SENT)
853 * If we get more than the supported amount of private data
854 * then we must fail this connection.
856 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
862 * copy the new data into our accumulation buffer.
864 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
866 ep->mpa_pkt_len += skb->len;
869 * if we don't even have the mpa message, then bail.
871 if (ep->mpa_pkt_len < sizeof(*mpa))
873 mpa = (struct mpa_message *) ep->mpa_pkt;
875 /* Validate MPA header. */
876 if (mpa->revision != mpa_rev) {
880 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
885 plen = ntohs(mpa->private_data_size);
888 * Fail if there's too much private data.
890 if (plen > MPA_MAX_PRIVATE_DATA) {
896 * If plen does not account for pkt size
898 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
903 ep->plen = (u8) plen;
906 * If we don't have all the pdata yet, then bail.
907 * We'll continue process when more data arrives.
909 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
912 if (mpa->flags & MPA_REJECT) {
918 * If we get here we have accumulated the entire mpa
919 * start reply message including private data. And
920 * the MPA header is valid.
922 state_set(&ep->com, FPDU_MODE);
923 ep->mpa_attr.initiator = 1;
924 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
925 ep->mpa_attr.recv_marker_enabled = markers_enabled;
926 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
927 ep->mpa_attr.version = mpa_rev;
928 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
929 "xmit_marker_enabled=%d, version=%d\n", __func__,
930 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
931 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
933 attrs.mpa_attr = ep->mpa_attr;
934 attrs.max_ird = ep->ird;
935 attrs.max_ord = ep->ord;
936 attrs.llp_stream_handle = ep;
937 attrs.next_state = IWCH_QP_STATE_RTS;
939 mask = IWCH_QP_ATTR_NEXT_STATE |
940 IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
941 IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
943 /* bind QP and TID with INIT_WR */
944 err = iwch_modify_qp(ep->com.qp->rhp,
945 ep->com.qp, mask, &attrs, 1);
949 if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
950 iwch_post_zb_read(ep->com.qp);
955 abort_connection(ep, skb, GFP_KERNEL);
957 connect_reply_upcall(ep, err);
961 static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
963 struct mpa_message *mpa;
966 PDBG("%s ep %p\n", __func__, ep);
969 * Stop mpa timer. If it expired, then the state has
970 * changed and we bail since ep_timeout already aborted
974 if (state_read(&ep->com) != MPA_REQ_WAIT)
978 * If we get more than the supported amount of private data
979 * then we must fail this connection.
981 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
982 abort_connection(ep, skb, GFP_KERNEL);
986 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
989 * Copy the new data into our accumulation buffer.
991 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
993 ep->mpa_pkt_len += skb->len;
996 * If we don't even have the mpa message, then bail.
997 * We'll continue process when more data arrives.
999 if (ep->mpa_pkt_len < sizeof(*mpa))
1001 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1002 mpa = (struct mpa_message *) ep->mpa_pkt;
1005 * Validate MPA Header.
1007 if (mpa->revision != mpa_rev) {
1008 abort_connection(ep, skb, GFP_KERNEL);
1012 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1013 abort_connection(ep, skb, GFP_KERNEL);
1017 plen = ntohs(mpa->private_data_size);
1020 * Fail if there's too much private data.
1022 if (plen > MPA_MAX_PRIVATE_DATA) {
1023 abort_connection(ep, skb, GFP_KERNEL);
1028 * If plen does not account for pkt size
1030 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1031 abort_connection(ep, skb, GFP_KERNEL);
1034 ep->plen = (u8) plen;
1037 * If we don't have all the pdata yet, then bail.
1039 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1043 * If we get here we have accumulated the entire mpa
1044 * start reply message including private data.
1046 ep->mpa_attr.initiator = 0;
1047 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1048 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1049 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1050 ep->mpa_attr.version = mpa_rev;
1051 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1052 "xmit_marker_enabled=%d, version=%d\n", __func__,
1053 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1054 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1056 state_set(&ep->com, MPA_REQ_RCVD);
1059 connect_request_upcall(ep);
1063 static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1065 struct iwch_ep *ep = ctx;
1066 struct cpl_rx_data *hdr = cplhdr(skb);
1067 unsigned int dlen = ntohs(hdr->len);
1069 PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1071 skb_pull(skb, sizeof(*hdr));
1072 skb_trim(skb, dlen);
1074 ep->rcv_seq += dlen;
1075 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1077 switch (state_read(&ep->com)) {
1079 process_mpa_reply(ep, skb);
1082 process_mpa_request(ep, skb);
1087 printk(KERN_ERR MOD "%s Unexpected streaming data."
1088 " ep %p state %d tid %d\n",
1089 __func__, ep, state_read(&ep->com), ep->hwtid);
1092 * The ep will timeout and inform the ULP of the failure.
1098 /* update RX credits */
1099 update_rx_credits(ep, dlen);
1101 return CPL_RET_BUF_DONE;
1105 * Upcall from the adapter indicating data has been transmitted.
1106 * For us its just the single MPA request or reply. We can now free
1107 * the skb holding the mpa message.
1109 static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1111 struct iwch_ep *ep = ctx;
1112 struct cpl_wr_ack *hdr = cplhdr(skb);
1113 unsigned int credits = ntohs(hdr->credits);
1115 PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1118 PDBG(KERN_ERR "%s 0 credit ack ep %p state %u\n",
1119 __func__, ep, state_read(&ep->com));
1120 return CPL_RET_BUF_DONE;
1123 BUG_ON(credits != 1);
1124 dst_confirm(ep->dst);
1126 PDBG("%s rdma_init wr_ack ep %p state %u\n",
1127 __func__, ep, state_read(&ep->com));
1128 if (ep->mpa_attr.initiator) {
1129 PDBG("%s initiator ep %p state %u\n",
1130 __func__, ep, state_read(&ep->com));
1132 iwch_post_zb_read(ep->com.qp);
1134 PDBG("%s responder ep %p state %u\n",
1135 __func__, ep, state_read(&ep->com));
1136 ep->com.rpl_done = 1;
1137 wake_up(&ep->com.waitq);
1140 PDBG("%s lsm ack ep %p state %u freeing skb\n",
1141 __func__, ep, state_read(&ep->com));
1142 kfree_skb(ep->mpa_skb);
1145 return CPL_RET_BUF_DONE;
1148 static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1150 struct iwch_ep *ep = ctx;
1151 unsigned long flags;
1154 PDBG("%s ep %p\n", __func__, ep);
1158 * We get 2 abort replies from the HW. The first one must
1159 * be ignored except for scribbling that we need one more.
1161 if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1162 return CPL_RET_BUF_DONE;
1165 spin_lock_irqsave(&ep->com.lock, flags);
1166 switch (ep->com.state) {
1168 close_complete_upcall(ep);
1169 __state_set(&ep->com, DEAD);
1173 printk(KERN_ERR "%s ep %p state %d\n",
1174 __func__, ep, ep->com.state);
1177 spin_unlock_irqrestore(&ep->com.lock, flags);
1180 release_ep_resources(ep);
1181 return CPL_RET_BUF_DONE;
1185 * Return whether a failed active open has allocated a TID
1187 static inline int act_open_has_tid(int status)
1189 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1190 status != CPL_ERR_ARP_MISS;
1193 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1195 struct iwch_ep *ep = ctx;
1196 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1198 PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1199 status2errno(rpl->status));
1200 connect_reply_upcall(ep, status2errno(rpl->status));
1201 state_set(&ep->com, DEAD);
1202 if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1203 release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1204 cxgb3_free_atid(ep->com.tdev, ep->atid);
1205 dst_release(ep->dst);
1206 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1208 return CPL_RET_BUF_DONE;
1211 static int listen_start(struct iwch_listen_ep *ep)
1213 struct sk_buff *skb;
1214 struct cpl_pass_open_req *req;
1216 PDBG("%s ep %p\n", __func__, ep);
1217 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1219 printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1223 req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1224 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1225 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1226 req->local_port = ep->com.local_addr.sin_port;
1227 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1230 req->peer_netmask = 0;
1231 req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1232 req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1233 req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1236 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1239 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1241 struct iwch_listen_ep *ep = ctx;
1242 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1244 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1245 rpl->status, status2errno(rpl->status));
1246 ep->com.rpl_err = status2errno(rpl->status);
1247 ep->com.rpl_done = 1;
1248 wake_up(&ep->com.waitq);
1250 return CPL_RET_BUF_DONE;
1253 static int listen_stop(struct iwch_listen_ep *ep)
1255 struct sk_buff *skb;
1256 struct cpl_close_listserv_req *req;
1258 PDBG("%s ep %p\n", __func__, ep);
1259 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1261 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1264 req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1265 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1267 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1269 return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1272 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1275 struct iwch_listen_ep *ep = ctx;
1276 struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1278 PDBG("%s ep %p\n", __func__, ep);
1279 ep->com.rpl_err = status2errno(rpl->status);
1280 ep->com.rpl_done = 1;
1281 wake_up(&ep->com.waitq);
1282 return CPL_RET_BUF_DONE;
1285 static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1287 struct cpl_pass_accept_rpl *rpl;
1288 unsigned int mtu_idx;
1289 u32 opt0h, opt0l, opt2;
1292 PDBG("%s ep %p\n", __func__, ep);
1293 BUG_ON(skb_cloned(skb));
1294 skb_trim(skb, sizeof(*rpl));
1296 mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1297 wscale = compute_wscale(rcv_win);
1298 opt0h = V_NAGLE(0) |
1302 V_WND_SCALE(wscale) |
1303 V_MSS_IDX(mtu_idx) |
1304 V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1305 opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1306 opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
1309 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1310 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1311 rpl->peer_ip = peer_ip;
1312 rpl->opt0h = htonl(opt0h);
1313 rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1314 rpl->opt2 = htonl(opt2);
1315 rpl->rsvd = rpl->opt2; /* workaround for HW bug */
1316 skb->priority = CPL_PRIORITY_SETUP;
1317 iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1322 static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1323 struct sk_buff *skb)
1325 PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1327 BUG_ON(skb_cloned(skb));
1328 skb_trim(skb, sizeof(struct cpl_tid_release));
1331 if (tdev->type != T3A)
1332 release_tid(tdev, hwtid, skb);
1334 struct cpl_pass_accept_rpl *rpl;
1337 skb->priority = CPL_PRIORITY_SETUP;
1338 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1339 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1341 rpl->peer_ip = peer_ip;
1342 rpl->opt0h = htonl(F_TCAM_BYPASS);
1343 rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1345 rpl->rsvd = rpl->opt2;
1346 iwch_cxgb3_ofld_send(tdev, skb);
1350 static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1352 struct iwch_ep *child_ep, *parent_ep = ctx;
1353 struct cpl_pass_accept_req *req = cplhdr(skb);
1354 unsigned int hwtid = GET_TID(req);
1355 struct dst_entry *dst;
1356 struct l2t_entry *l2t;
1360 PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1362 if (state_read(&parent_ep->com) != LISTEN) {
1363 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1369 * Find the netdev for this connection request.
1371 tim.mac_addr = req->dst_mac;
1372 tim.vlan_tag = ntohs(req->vlan_tag);
1373 if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1374 printk(KERN_ERR "%s bad dst mac %pM\n",
1375 __func__, req->dst_mac);
1379 /* Find output route */
1380 rt = find_route(tdev,
1384 req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1386 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1391 l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1393 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1398 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1400 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1402 l2t_release(L2DATA(tdev), l2t);
1406 state_set(&child_ep->com, CONNECTING);
1407 child_ep->com.tdev = tdev;
1408 child_ep->com.cm_id = NULL;
1409 child_ep->com.local_addr.sin_family = PF_INET;
1410 child_ep->com.local_addr.sin_port = req->local_port;
1411 child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1412 child_ep->com.remote_addr.sin_family = PF_INET;
1413 child_ep->com.remote_addr.sin_port = req->peer_port;
1414 child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1415 get_ep(&parent_ep->com);
1416 child_ep->parent_ep = parent_ep;
1417 child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1418 child_ep->l2t = l2t;
1419 child_ep->dst = dst;
1420 child_ep->hwtid = hwtid;
1421 init_timer(&child_ep->timer);
1422 cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1423 accept_cr(child_ep, req->peer_ip, skb);
1426 reject_cr(tdev, hwtid, req->peer_ip, skb);
1428 return CPL_RET_BUF_DONE;
1431 static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1433 struct iwch_ep *ep = ctx;
1434 struct cpl_pass_establish *req = cplhdr(skb);
1436 PDBG("%s ep %p\n", __func__, ep);
1437 ep->snd_seq = ntohl(req->snd_isn);
1438 ep->rcv_seq = ntohl(req->rcv_isn);
1440 set_emss(ep, ntohs(req->tcp_opt));
1442 dst_confirm(ep->dst);
1443 state_set(&ep->com, MPA_REQ_WAIT);
1446 return CPL_RET_BUF_DONE;
1449 static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1451 struct iwch_ep *ep = ctx;
1452 struct iwch_qp_attributes attrs;
1453 unsigned long flags;
1457 PDBG("%s ep %p\n", __func__, ep);
1458 dst_confirm(ep->dst);
1460 spin_lock_irqsave(&ep->com.lock, flags);
1461 switch (ep->com.state) {
1463 __state_set(&ep->com, CLOSING);
1466 __state_set(&ep->com, CLOSING);
1467 connect_reply_upcall(ep, -ECONNRESET);
1472 * We're gonna mark this puppy DEAD, but keep
1473 * the reference on it until the ULP accepts or
1474 * rejects the CR. Also wake up anyone waiting
1475 * in rdma connection migration (see iwch_accept_cr()).
1477 __state_set(&ep->com, CLOSING);
1478 ep->com.rpl_done = 1;
1479 ep->com.rpl_err = -ECONNRESET;
1480 PDBG("waking up ep %p\n", ep);
1481 wake_up(&ep->com.waitq);
1484 __state_set(&ep->com, CLOSING);
1485 ep->com.rpl_done = 1;
1486 ep->com.rpl_err = -ECONNRESET;
1487 PDBG("waking up ep %p\n", ep);
1488 wake_up(&ep->com.waitq);
1492 __state_set(&ep->com, CLOSING);
1493 attrs.next_state = IWCH_QP_STATE_CLOSING;
1494 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1495 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1496 peer_close_upcall(ep);
1502 __state_set(&ep->com, MORIBUND);
1507 if (ep->com.cm_id && ep->com.qp) {
1508 attrs.next_state = IWCH_QP_STATE_IDLE;
1509 iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1510 IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1512 close_complete_upcall(ep);
1513 __state_set(&ep->com, DEAD);
1523 spin_unlock_irqrestore(&ep->com.lock, flags);
1525 iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1527 release_ep_resources(ep);
1528 return CPL_RET_BUF_DONE;
1532 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1534 static int is_neg_adv_abort(unsigned int status)
1536 return status == CPL_ERR_RTX_NEG_ADVICE ||
1537 status == CPL_ERR_PERSIST_NEG_ADVICE;
1540 static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1542 struct cpl_abort_req_rss *req = cplhdr(skb);
1543 struct iwch_ep *ep = ctx;
1544 struct cpl_abort_rpl *rpl;
1545 struct sk_buff *rpl_skb;
1546 struct iwch_qp_attributes attrs;
1549 unsigned long flags;
1551 if (is_neg_adv_abort(req->status)) {
1552 PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1554 t3_l2t_send_event(ep->com.tdev, ep->l2t);
1555 return CPL_RET_BUF_DONE;
1559 * We get 2 peer aborts from the HW. The first one must
1560 * be ignored except for scribbling that we need one more.
1562 if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
1563 return CPL_RET_BUF_DONE;
1566 spin_lock_irqsave(&ep->com.lock, flags);
1567 PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1568 switch (ep->com.state) {
1576 connect_reply_upcall(ep, -ECONNRESET);
1579 ep->com.rpl_done = 1;
1580 ep->com.rpl_err = -ECONNRESET;
1581 PDBG("waking up ep %p\n", ep);
1582 wake_up(&ep->com.waitq);
1587 * We're gonna mark this puppy DEAD, but keep
1588 * the reference on it until the ULP accepts or
1589 * rejects the CR. Also wake up anyone waiting
1590 * in rdma connection migration (see iwch_accept_cr()).
1592 ep->com.rpl_done = 1;
1593 ep->com.rpl_err = -ECONNRESET;
1594 PDBG("waking up ep %p\n", ep);
1595 wake_up(&ep->com.waitq);
1602 if (ep->com.cm_id && ep->com.qp) {
1603 attrs.next_state = IWCH_QP_STATE_ERROR;
1604 ret = iwch_modify_qp(ep->com.qp->rhp,
1605 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1609 "%s - qp <- error failed!\n",
1612 peer_abort_upcall(ep);
1617 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1618 spin_unlock_irqrestore(&ep->com.lock, flags);
1619 return CPL_RET_BUF_DONE;
1624 dst_confirm(ep->dst);
1625 if (ep->com.state != ABORTING) {
1626 __state_set(&ep->com, DEAD);
1629 spin_unlock_irqrestore(&ep->com.lock, flags);
1631 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1633 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1638 rpl_skb->priority = CPL_PRIORITY_DATA;
1639 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1640 rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1641 rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1642 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1643 rpl->cmd = CPL_ABORT_NO_RST;
1644 iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1647 release_ep_resources(ep);
1648 return CPL_RET_BUF_DONE;
1651 static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1653 struct iwch_ep *ep = ctx;
1654 struct iwch_qp_attributes attrs;
1655 unsigned long flags;
1658 PDBG("%s ep %p\n", __func__, ep);
1661 /* The cm_id may be null if we failed to connect */
1662 spin_lock_irqsave(&ep->com.lock, flags);
1663 switch (ep->com.state) {
1665 __state_set(&ep->com, MORIBUND);
1669 if ((ep->com.cm_id) && (ep->com.qp)) {
1670 attrs.next_state = IWCH_QP_STATE_IDLE;
1671 iwch_modify_qp(ep->com.qp->rhp,
1673 IWCH_QP_ATTR_NEXT_STATE,
1676 close_complete_upcall(ep);
1677 __state_set(&ep->com, DEAD);
1687 spin_unlock_irqrestore(&ep->com.lock, flags);
1689 release_ep_resources(ep);
1690 return CPL_RET_BUF_DONE;
1694 * T3A does 3 things when a TERM is received:
1695 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1696 * 2) generate an async event on the QP with the TERMINATE opcode
1697 * 3) post a TERMINATE opcde cqe into the associated CQ.
1699 * For (1), we save the message in the qp for later consumer consumption.
1700 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1701 * For (3), we toss the CQE in cxio_poll_cq().
1703 * terminate() handles case (1)...
1705 static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1707 struct iwch_ep *ep = ctx;
1709 if (state_read(&ep->com) != FPDU_MODE)
1710 return CPL_RET_BUF_DONE;
1712 PDBG("%s ep %p\n", __func__, ep);
1713 skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1714 PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1715 skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1717 ep->com.qp->attr.terminate_msg_len = skb->len;
1718 ep->com.qp->attr.is_terminate_local = 0;
1719 return CPL_RET_BUF_DONE;
1722 static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1724 struct cpl_rdma_ec_status *rep = cplhdr(skb);
1725 struct iwch_ep *ep = ctx;
1727 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
1730 struct iwch_qp_attributes attrs;
1732 printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1733 __func__, ep->hwtid);
1735 attrs.next_state = IWCH_QP_STATE_ERROR;
1736 iwch_modify_qp(ep->com.qp->rhp,
1737 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1739 abort_connection(ep, NULL, GFP_KERNEL);
1741 return CPL_RET_BUF_DONE;
1744 static void ep_timeout(unsigned long arg)
1746 struct iwch_ep *ep = (struct iwch_ep *)arg;
1747 struct iwch_qp_attributes attrs;
1748 unsigned long flags;
1751 spin_lock_irqsave(&ep->com.lock, flags);
1752 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
1754 switch (ep->com.state) {
1756 __state_set(&ep->com, ABORTING);
1757 connect_reply_upcall(ep, -ETIMEDOUT);
1760 __state_set(&ep->com, ABORTING);
1764 if (ep->com.cm_id && ep->com.qp) {
1765 attrs.next_state = IWCH_QP_STATE_ERROR;
1766 iwch_modify_qp(ep->com.qp->rhp,
1767 ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1770 __state_set(&ep->com, ABORTING);
1773 printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1774 __func__, ep, ep->com.state);
1778 spin_unlock_irqrestore(&ep->com.lock, flags);
1780 abort_connection(ep, NULL, GFP_ATOMIC);
1784 int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1787 struct iwch_ep *ep = to_ep(cm_id);
1788 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1790 if (state_read(&ep->com) == DEAD) {
1794 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1796 abort_connection(ep, NULL, GFP_KERNEL);
1798 err = send_mpa_reject(ep, pdata, pdata_len);
1799 err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1805 int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1808 struct iwch_qp_attributes attrs;
1809 enum iwch_qp_attr_mask mask;
1810 struct iwch_ep *ep = to_ep(cm_id);
1811 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1812 struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1814 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1815 if (state_read(&ep->com) == DEAD) {
1820 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1823 if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1824 (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1825 abort_connection(ep, NULL, GFP_KERNEL);
1830 cm_id->add_ref(cm_id);
1831 ep->com.cm_id = cm_id;
1834 ep->ird = conn_param->ird;
1835 ep->ord = conn_param->ord;
1837 if (peer2peer && ep->ird == 0)
1840 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1842 /* bind QP to EP and move to RTS */
1843 attrs.mpa_attr = ep->mpa_attr;
1844 attrs.max_ird = ep->ird;
1845 attrs.max_ord = ep->ord;
1846 attrs.llp_stream_handle = ep;
1847 attrs.next_state = IWCH_QP_STATE_RTS;
1849 /* bind QP and TID with INIT_WR */
1850 mask = IWCH_QP_ATTR_NEXT_STATE |
1851 IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1852 IWCH_QP_ATTR_MPA_ATTR |
1853 IWCH_QP_ATTR_MAX_IRD |
1854 IWCH_QP_ATTR_MAX_ORD;
1856 err = iwch_modify_qp(ep->com.qp->rhp,
1857 ep->com.qp, mask, &attrs, 1);
1861 /* if needed, wait for wr_ack */
1862 if (iwch_rqes_posted(qp)) {
1863 wait_event(ep->com.waitq, ep->com.rpl_done);
1864 err = ep->com.rpl_err;
1869 err = send_mpa_reply(ep, conn_param->private_data,
1870 conn_param->private_data_len);
1875 state_set(&ep->com, FPDU_MODE);
1876 established_upcall(ep);
1880 ep->com.cm_id = NULL;
1882 cm_id->rem_ref(cm_id);
1888 static int is_loopback_dst(struct iw_cm_id *cm_id)
1890 struct net_device *dev;
1892 dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1899 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1902 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1906 if (is_loopback_dst(cm_id)) {
1911 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1913 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1917 init_timer(&ep->timer);
1918 ep->plen = conn_param->private_data_len;
1920 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1921 conn_param->private_data, ep->plen);
1922 ep->ird = conn_param->ird;
1923 ep->ord = conn_param->ord;
1925 if (peer2peer && ep->ord == 0)
1928 ep->com.tdev = h->rdev.t3cdev_p;
1930 cm_id->add_ref(cm_id);
1931 ep->com.cm_id = cm_id;
1932 ep->com.qp = get_qhp(h, conn_param->qpn);
1933 BUG_ON(!ep->com.qp);
1934 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1938 * Allocate an active TID to initiate a TCP connection.
1940 ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1941 if (ep->atid == -1) {
1942 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1948 rt = find_route(h->rdev.t3cdev_p,
1949 cm_id->local_addr.sin_addr.s_addr,
1950 cm_id->remote_addr.sin_addr.s_addr,
1951 cm_id->local_addr.sin_port,
1952 cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1954 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1955 err = -EHOSTUNREACH;
1958 ep->dst = &rt->u.dst;
1960 /* get a l2t entry */
1961 ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1962 ep->dst->neighbour->dev);
1964 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1969 state_set(&ep->com, CONNECTING);
1970 ep->tos = IPTOS_LOWDELAY;
1971 ep->com.local_addr = cm_id->local_addr;
1972 ep->com.remote_addr = cm_id->remote_addr;
1974 /* send connect request to rnic */
1975 err = send_connect(ep);
1979 l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1981 dst_release(ep->dst);
1983 cxgb3_free_atid(ep->com.tdev, ep->atid);
1985 cm_id->rem_ref(cm_id);
1991 int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1994 struct iwch_dev *h = to_iwch_dev(cm_id->device);
1995 struct iwch_listen_ep *ep;
2000 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2002 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2006 PDBG("%s ep %p\n", __func__, ep);
2007 ep->com.tdev = h->rdev.t3cdev_p;
2008 cm_id->add_ref(cm_id);
2009 ep->com.cm_id = cm_id;
2010 ep->backlog = backlog;
2011 ep->com.local_addr = cm_id->local_addr;
2014 * Allocate a server TID.
2016 ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
2017 if (ep->stid == -1) {
2018 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2023 state_set(&ep->com, LISTEN);
2024 err = listen_start(ep);
2028 /* wait for pass_open_rpl */
2029 wait_event(ep->com.waitq, ep->com.rpl_done);
2030 err = ep->com.rpl_err;
2032 cm_id->provider_data = ep;
2036 cxgb3_free_stid(ep->com.tdev, ep->stid);
2038 cm_id->rem_ref(cm_id);
2045 int iwch_destroy_listen(struct iw_cm_id *cm_id)
2048 struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2050 PDBG("%s ep %p\n", __func__, ep);
2053 state_set(&ep->com, DEAD);
2054 ep->com.rpl_done = 0;
2055 ep->com.rpl_err = 0;
2056 err = listen_stop(ep);
2059 wait_event(ep->com.waitq, ep->com.rpl_done);
2060 cxgb3_free_stid(ep->com.tdev, ep->stid);
2062 err = ep->com.rpl_err;
2063 cm_id->rem_ref(cm_id);
2068 int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2071 unsigned long flags;
2074 struct t3cdev *tdev;
2075 struct cxio_rdev *rdev;
2077 spin_lock_irqsave(&ep->com.lock, flags);
2079 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2080 states[ep->com.state], abrupt);
2082 tdev = (struct t3cdev *)ep->com.tdev;
2083 rdev = (struct cxio_rdev *)tdev->ulp;
2084 if (cxio_fatal_error(rdev)) {
2086 close_complete_upcall(ep);
2087 ep->com.state = DEAD;
2089 switch (ep->com.state) {
2097 ep->com.state = ABORTING;
2099 ep->com.state = CLOSING;
2102 set_bit(CLOSE_SENT, &ep->com.flags);
2105 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2109 ep->com.state = ABORTING;
2111 ep->com.state = MORIBUND;
2117 PDBG("%s ignoring disconnect ep %p state %u\n",
2118 __func__, ep, ep->com.state);
2125 spin_unlock_irqrestore(&ep->com.lock, flags);
2128 ret = send_abort(ep, NULL, gfp);
2130 ret = send_halfclose(ep, gfp);
2135 release_ep_resources(ep);
2139 int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2140 struct l2t_entry *l2t)
2142 struct iwch_ep *ep = ctx;
2147 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2150 l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2158 * All the CM events are handled on a work queue to have a safe context.
2160 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2162 struct iwch_ep_common *epc = ctx;
2167 * Save ctx and tdev in the skb->cb area.
2169 *((void **) skb->cb) = ctx;
2170 *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2173 * Queue the skb and schedule the worker thread.
2175 skb_queue_tail(&rxq, skb);
2176 queue_work(workq, &skb_work);
2180 static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2182 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2184 if (rpl->status != CPL_ERR_NONE) {
2185 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2186 "for tid %u\n", rpl->status, GET_TID(rpl));
2188 return CPL_RET_BUF_DONE;
2191 int __init iwch_cm_init(void)
2193 skb_queue_head_init(&rxq);
2195 workq = create_singlethread_workqueue("iw_cxgb3");
2200 * All upcalls from the T3 Core go to sched() to
2201 * schedule the processing on a work queue.
2203 t3c_handlers[CPL_ACT_ESTABLISH] = sched;
2204 t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
2205 t3c_handlers[CPL_RX_DATA] = sched;
2206 t3c_handlers[CPL_TX_DMA_ACK] = sched;
2207 t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
2208 t3c_handlers[CPL_ABORT_RPL] = sched;
2209 t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
2210 t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
2211 t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
2212 t3c_handlers[CPL_PASS_ESTABLISH] = sched;
2213 t3c_handlers[CPL_PEER_CLOSE] = sched;
2214 t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
2215 t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
2216 t3c_handlers[CPL_RDMA_TERMINATE] = sched;
2217 t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
2218 t3c_handlers[CPL_SET_TCB_RPL] = set_tcb_rpl;
2221 * These are the real handlers that are called from a
2224 work_handlers[CPL_ACT_ESTABLISH] = act_establish;
2225 work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
2226 work_handlers[CPL_RX_DATA] = rx_data;
2227 work_handlers[CPL_TX_DMA_ACK] = tx_ack;
2228 work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
2229 work_handlers[CPL_ABORT_RPL] = abort_rpl;
2230 work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
2231 work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
2232 work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
2233 work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
2234 work_handlers[CPL_PEER_CLOSE] = peer_close;
2235 work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
2236 work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
2237 work_handlers[CPL_RDMA_TERMINATE] = terminate;
2238 work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
2242 void __exit iwch_cm_term(void)
2244 flush_workqueue(workq);
2245 destroy_workqueue(workq);