2 * Copyright (c) 2009-2014 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 <linux/tcp.h>
41 #include <linux/if_vlan.h>
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
50 #include <rdma/ib_addr.h>
52 #include <libcxgb_cm.h>
56 static char *states[] = {
73 module_param(nocong, int, 0644);
74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
76 static int enable_ecn;
77 module_param(enable_ecn, int, 0644);
78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
80 static int dack_mode = 1;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
84 uint c4iw_max_read_depth = 32;
85 module_param(c4iw_max_read_depth, int, 0644);
86 MODULE_PARM_DESC(c4iw_max_read_depth,
87 "Per-connection max ORD/IRD (default=32)");
89 static int enable_tcp_timestamps;
90 module_param(enable_tcp_timestamps, int, 0644);
91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
93 static int enable_tcp_sack;
94 module_param(enable_tcp_sack, int, 0644);
95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
97 static int enable_tcp_window_scaling = 1;
98 module_param(enable_tcp_window_scaling, int, 0644);
99 MODULE_PARM_DESC(enable_tcp_window_scaling,
100 "Enable tcp window scaling (default=1)");
103 module_param(c4iw_debug, int, 0644);
104 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
106 static int peer2peer = 1;
107 module_param(peer2peer, int, 0644);
108 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
110 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
111 module_param(p2p_type, int, 0644);
112 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
113 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
115 static int ep_timeout_secs = 60;
116 module_param(ep_timeout_secs, int, 0644);
117 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
118 "in seconds (default=60)");
120 static int mpa_rev = 2;
121 module_param(mpa_rev, int, 0644);
122 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
123 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
124 " compliant (default=2)");
126 static int markers_enabled;
127 module_param(markers_enabled, int, 0644);
128 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
130 static int crc_enabled = 1;
131 module_param(crc_enabled, int, 0644);
132 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
134 static int rcv_win = 256 * 1024;
135 module_param(rcv_win, int, 0644);
136 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
138 static int snd_win = 128 * 1024;
139 module_param(snd_win, int, 0644);
140 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
142 static struct workqueue_struct *workq;
144 static struct sk_buff_head rxq;
146 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
147 static void ep_timeout(unsigned long arg);
148 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
149 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
151 static LIST_HEAD(timeout_list);
152 static spinlock_t timeout_lock;
154 static void deref_cm_id(struct c4iw_ep_common *epc)
156 epc->cm_id->rem_ref(epc->cm_id);
158 set_bit(CM_ID_DEREFED, &epc->history);
161 static void ref_cm_id(struct c4iw_ep_common *epc)
163 set_bit(CM_ID_REFED, &epc->history);
164 epc->cm_id->add_ref(epc->cm_id);
167 static void deref_qp(struct c4iw_ep *ep)
169 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
170 clear_bit(QP_REFERENCED, &ep->com.flags);
171 set_bit(QP_DEREFED, &ep->com.history);
174 static void ref_qp(struct c4iw_ep *ep)
176 set_bit(QP_REFERENCED, &ep->com.flags);
177 set_bit(QP_REFED, &ep->com.history);
178 c4iw_qp_add_ref(&ep->com.qp->ibqp);
181 static void start_ep_timer(struct c4iw_ep *ep)
183 PDBG("%s ep %p\n", __func__, ep);
184 if (timer_pending(&ep->timer)) {
185 pr_err("%s timer already started! ep %p\n",
189 clear_bit(TIMEOUT, &ep->com.flags);
190 c4iw_get_ep(&ep->com);
191 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
192 ep->timer.data = (unsigned long)ep;
193 ep->timer.function = ep_timeout;
194 add_timer(&ep->timer);
197 static int stop_ep_timer(struct c4iw_ep *ep)
199 PDBG("%s ep %p stopping\n", __func__, ep);
200 del_timer_sync(&ep->timer);
201 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
202 c4iw_put_ep(&ep->com);
208 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
209 struct l2t_entry *l2e)
213 if (c4iw_fatal_error(rdev)) {
215 PDBG("%s - device in error state - dropping\n", __func__);
218 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
221 else if (error == NET_XMIT_DROP)
223 return error < 0 ? error : 0;
226 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
230 if (c4iw_fatal_error(rdev)) {
232 PDBG("%s - device in error state - dropping\n", __func__);
235 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
238 return error < 0 ? error : 0;
241 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
243 u32 len = roundup(sizeof(struct cpl_tid_release), 16);
245 skb = get_skb(skb, len, GFP_KERNEL);
249 cxgb_mk_tid_release(skb, len, hwtid, 0);
250 c4iw_ofld_send(rdev, skb);
254 static void set_emss(struct c4iw_ep *ep, u16 opt)
256 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
257 ((AF_INET == ep->com.remote_addr.ss_family) ?
258 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
259 sizeof(struct tcphdr);
261 if (TCPOPT_TSTAMP_G(opt))
262 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
266 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
267 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
268 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
272 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
274 enum c4iw_ep_state state;
276 mutex_lock(&epc->mutex);
278 mutex_unlock(&epc->mutex);
282 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
287 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
289 mutex_lock(&epc->mutex);
290 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
291 __state_set(epc, new);
292 mutex_unlock(&epc->mutex);
296 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
302 len = roundup(sizeof(union cpl_wr_size), 16);
303 for (i = 0; i < size; i++) {
304 skb = alloc_skb(len, GFP_KERNEL);
307 skb_queue_tail(ep_skb_list, skb);
311 skb_queue_purge(ep_skb_list);
315 static void *alloc_ep(int size, gfp_t gfp)
317 struct c4iw_ep_common *epc;
319 epc = kzalloc(size, gfp);
321 kref_init(&epc->kref);
322 mutex_init(&epc->mutex);
323 c4iw_init_wr_wait(&epc->wr_wait);
325 PDBG("%s alloc ep %p\n", __func__, epc);
329 static void remove_ep_tid(struct c4iw_ep *ep)
333 spin_lock_irqsave(&ep->com.dev->lock, flags);
334 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
335 if (idr_is_empty(&ep->com.dev->hwtid_idr))
336 wake_up(&ep->com.dev->wait);
337 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
340 static void insert_ep_tid(struct c4iw_ep *ep)
344 spin_lock_irqsave(&ep->com.dev->lock, flags);
345 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
346 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
350 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
352 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
357 spin_lock_irqsave(&dev->lock, flags);
358 ep = idr_find(&dev->hwtid_idr, tid);
360 c4iw_get_ep(&ep->com);
361 spin_unlock_irqrestore(&dev->lock, flags);
366 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
368 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
371 struct c4iw_listen_ep *ep;
374 spin_lock_irqsave(&dev->lock, flags);
375 ep = idr_find(&dev->stid_idr, stid);
377 c4iw_get_ep(&ep->com);
378 spin_unlock_irqrestore(&dev->lock, flags);
382 void _c4iw_free_ep(struct kref *kref)
386 ep = container_of(kref, struct c4iw_ep, com.kref);
387 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
388 if (test_bit(QP_REFERENCED, &ep->com.flags))
390 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
391 if (ep->com.remote_addr.ss_family == AF_INET6) {
392 struct sockaddr_in6 *sin6 =
393 (struct sockaddr_in6 *)
397 ep->com.dev->rdev.lldi.ports[0],
398 (const u32 *)&sin6->sin6_addr.s6_addr,
401 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
402 dst_release(ep->dst);
403 cxgb4_l2t_release(ep->l2t);
405 kfree_skb(ep->mpa_skb);
407 if (!skb_queue_empty(&ep->com.ep_skb_list))
408 skb_queue_purge(&ep->com.ep_skb_list);
412 static void release_ep_resources(struct c4iw_ep *ep)
414 set_bit(RELEASE_RESOURCES, &ep->com.flags);
417 * If we have a hwtid, then remove it from the idr table
418 * so lookups will no longer find this endpoint. Otherwise
419 * we have a race where one thread finds the ep ptr just
420 * before the other thread is freeing the ep memory.
424 c4iw_put_ep(&ep->com);
427 static int status2errno(int status)
432 case CPL_ERR_CONN_RESET:
434 case CPL_ERR_ARP_MISS:
435 return -EHOSTUNREACH;
436 case CPL_ERR_CONN_TIMEDOUT:
438 case CPL_ERR_TCAM_FULL:
440 case CPL_ERR_CONN_EXIST:
448 * Try and reuse skbs already allocated...
450 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
452 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
455 skb_reset_transport_header(skb);
457 skb = alloc_skb(len, gfp);
459 t4_set_arp_err_handler(skb, NULL, NULL);
463 static struct net_device *get_real_dev(struct net_device *egress_dev)
465 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
468 static void arp_failure_discard(void *handle, struct sk_buff *skb)
470 pr_err(MOD "ARP failure\n");
474 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
476 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
481 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
482 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
485 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
489 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
490 release_ep_resources(ep);
494 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
498 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
499 c4iw_put_ep(&ep->parent_ep->com);
500 release_ep_resources(ep);
505 * Fake up a special CPL opcode and call sched() so process_work() will call
506 * _put_ep_safe() in a safe context to free the ep resources. This is needed
507 * because ARP error handlers are called in an ATOMIC context, and
508 * _c4iw_free_ep() needs to block.
510 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
513 struct cpl_act_establish *rpl = cplhdr(skb);
515 /* Set our special ARP_FAILURE opcode */
516 rpl->ot.opcode = cpl;
519 * Save ep in the skb->cb area, after where sched() will save the dev
522 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
523 sched(ep->com.dev, skb);
526 /* Handle an ARP failure for an accept */
527 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
529 struct c4iw_ep *ep = handle;
531 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
534 __state_set(&ep->com, DEAD);
535 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
539 * Handle an ARP failure for an active open.
541 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
543 struct c4iw_ep *ep = handle;
545 printk(KERN_ERR MOD "ARP failure during connect\n");
546 connect_reply_upcall(ep, -EHOSTUNREACH);
547 __state_set(&ep->com, DEAD);
548 if (ep->com.remote_addr.ss_family == AF_INET6) {
549 struct sockaddr_in6 *sin6 =
550 (struct sockaddr_in6 *)&ep->com.local_addr;
551 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
552 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
554 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
555 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
556 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
560 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
563 static void abort_arp_failure(void *handle, struct sk_buff *skb)
566 struct c4iw_ep *ep = handle;
567 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
568 struct cpl_abort_req *req = cplhdr(skb);
570 PDBG("%s rdev %p\n", __func__, rdev);
571 req->cmd = CPL_ABORT_NO_RST;
572 ret = c4iw_ofld_send(rdev, skb);
574 __state_set(&ep->com, DEAD);
575 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
579 static int send_flowc(struct c4iw_ep *ep)
581 struct fw_flowc_wr *flowc;
582 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
584 u16 vlan = ep->l2t->vlan;
590 if (vlan == CPL_L2T_VLAN_NONE)
595 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
597 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
598 FW_FLOWC_WR_NPARAMS_V(nparams));
599 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
600 16)) | FW_WR_FLOWID_V(ep->hwtid));
602 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
603 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
604 (ep->com.dev->rdev.lldi.pf));
605 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
606 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
607 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
608 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
609 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
610 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
611 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
612 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
613 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
614 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
615 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
616 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
617 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
618 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
622 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
623 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
624 flowc->mnemval[8].val = cpu_to_be32(pri);
626 /* Pad WR to 16 byte boundary */
627 flowc->mnemval[8].mnemonic = 0;
628 flowc->mnemval[8].val = 0;
630 for (i = 0; i < 9; i++) {
631 flowc->mnemval[i].r4[0] = 0;
632 flowc->mnemval[i].r4[1] = 0;
633 flowc->mnemval[i].r4[2] = 0;
636 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
637 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
640 static int send_halfclose(struct c4iw_ep *ep)
642 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
643 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
645 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
649 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
650 NULL, arp_failure_discard);
652 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
655 static int send_abort(struct c4iw_ep *ep)
657 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
658 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
660 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
661 if (WARN_ON(!req_skb))
664 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
665 ep, abort_arp_failure);
667 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
670 static int send_connect(struct c4iw_ep *ep)
672 struct cpl_act_open_req *req = NULL;
673 struct cpl_t5_act_open_req *t5req = NULL;
674 struct cpl_t6_act_open_req *t6req = NULL;
675 struct cpl_act_open_req6 *req6 = NULL;
676 struct cpl_t5_act_open_req6 *t5req6 = NULL;
677 struct cpl_t6_act_open_req6 *t6req6 = NULL;
681 unsigned int mtu_idx;
683 int win, sizev4, sizev6, wrlen;
684 struct sockaddr_in *la = (struct sockaddr_in *)
686 struct sockaddr_in *ra = (struct sockaddr_in *)
687 &ep->com.remote_addr;
688 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
690 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
691 &ep->com.remote_addr;
693 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
694 u32 isn = (prandom_u32() & ~7UL) - 1;
695 struct net_device *netdev;
698 netdev = ep->com.dev->rdev.lldi.ports[0];
700 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
702 sizev4 = sizeof(struct cpl_act_open_req);
703 sizev6 = sizeof(struct cpl_act_open_req6);
706 sizev4 = sizeof(struct cpl_t5_act_open_req);
707 sizev6 = sizeof(struct cpl_t5_act_open_req6);
710 sizev4 = sizeof(struct cpl_t6_act_open_req);
711 sizev6 = sizeof(struct cpl_t6_act_open_req6);
714 pr_err("T%d Chip is not supported\n",
715 CHELSIO_CHIP_VERSION(adapter_type));
719 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
720 roundup(sizev4, 16) :
723 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
725 skb = get_skb(NULL, wrlen, GFP_KERNEL);
727 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
731 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
733 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
734 enable_tcp_timestamps,
735 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
736 wscale = cxgb_compute_wscale(rcv_win);
739 * Specify the largest window that will fit in opt0. The
740 * remainder will be specified in the rx_data_ack.
742 win = ep->rcv_win >> 10;
743 if (win > RCV_BUFSIZ_M)
746 opt0 = (nocong ? NO_CONG_F : 0) |
749 WND_SCALE_V(wscale) |
751 L2T_IDX_V(ep->l2t->idx) |
752 TX_CHAN_V(ep->tx_chan) |
753 SMAC_SEL_V(ep->smac_idx) |
754 DSCP_V(ep->tos >> 2) |
755 ULP_MODE_V(ULP_MODE_TCPDDP) |
757 opt2 = RX_CHANNEL_V(0) |
758 CCTRL_ECN_V(enable_ecn) |
759 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
760 if (enable_tcp_timestamps)
761 opt2 |= TSTAMPS_EN_F;
764 if (wscale && enable_tcp_window_scaling)
765 opt2 |= WND_SCALE_EN_F;
766 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
770 opt2 |= T5_OPT_2_VALID_F;
771 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
775 params = cxgb4_select_ntuple(netdev, ep->l2t);
777 if (ep->com.remote_addr.ss_family == AF_INET6)
778 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
779 (const u32 *)&la6->sin6_addr.s6_addr, 1);
781 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
783 if (ep->com.remote_addr.ss_family == AF_INET) {
784 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
786 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
790 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
792 INIT_TP_WR(t5req, 0);
793 req = (struct cpl_act_open_req *)t5req;
796 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
798 INIT_TP_WR(t6req, 0);
799 req = (struct cpl_act_open_req *)t6req;
800 t5req = (struct cpl_t5_act_open_req *)t6req;
803 pr_err("T%d Chip is not supported\n",
804 CHELSIO_CHIP_VERSION(adapter_type));
809 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
810 ((ep->rss_qid<<14) | ep->atid)));
811 req->local_port = la->sin_port;
812 req->peer_port = ra->sin_port;
813 req->local_ip = la->sin_addr.s_addr;
814 req->peer_ip = ra->sin_addr.s_addr;
815 req->opt0 = cpu_to_be64(opt0);
817 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
818 req->params = cpu_to_be32(params);
819 req->opt2 = cpu_to_be32(opt2);
821 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
823 cpu_to_be64(FILTER_TUPLE_V(params));
824 t5req->rsvd = cpu_to_be32(isn);
825 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
826 t5req->opt2 = cpu_to_be32(opt2);
829 cpu_to_be64(FILTER_TUPLE_V(params));
830 t6req->rsvd = cpu_to_be32(isn);
831 PDBG("%s snd_isn %u\n", __func__, t6req->rsvd);
832 t6req->opt2 = cpu_to_be32(opt2);
836 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
838 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
842 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
844 INIT_TP_WR(t5req6, 0);
845 req6 = (struct cpl_act_open_req6 *)t5req6;
848 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
850 INIT_TP_WR(t6req6, 0);
851 req6 = (struct cpl_act_open_req6 *)t6req6;
852 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
855 pr_err("T%d Chip is not supported\n",
856 CHELSIO_CHIP_VERSION(adapter_type));
861 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
862 ((ep->rss_qid<<14)|ep->atid)));
863 req6->local_port = la6->sin6_port;
864 req6->peer_port = ra6->sin6_port;
865 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
866 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
867 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
868 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
869 req6->opt0 = cpu_to_be64(opt0);
871 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
872 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
874 req6->opt2 = cpu_to_be32(opt2);
876 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
878 cpu_to_be64(FILTER_TUPLE_V(params));
879 t5req6->rsvd = cpu_to_be32(isn);
880 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
881 t5req6->opt2 = cpu_to_be32(opt2);
884 cpu_to_be64(FILTER_TUPLE_V(params));
885 t6req6->rsvd = cpu_to_be32(isn);
886 PDBG("%s snd_isn %u\n", __func__, t6req6->rsvd);
887 t6req6->opt2 = cpu_to_be32(opt2);
893 set_bit(ACT_OPEN_REQ, &ep->com.history);
894 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
896 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
897 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
898 (const u32 *)&la6->sin6_addr.s6_addr, 1);
902 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
905 int mpalen, wrlen, ret;
906 struct fw_ofld_tx_data_wr *req;
907 struct mpa_message *mpa;
908 struct mpa_v2_conn_params mpa_v2_params;
910 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
912 BUG_ON(skb_cloned(skb));
914 mpalen = sizeof(*mpa) + ep->plen;
915 if (mpa_rev_to_use == 2)
916 mpalen += sizeof(struct mpa_v2_conn_params);
917 wrlen = roundup(mpalen + sizeof *req, 16);
918 skb = get_skb(skb, wrlen, GFP_KERNEL);
920 connect_reply_upcall(ep, -ENOMEM);
923 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
925 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
926 memset(req, 0, wrlen);
927 req->op_to_immdlen = cpu_to_be32(
928 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
930 FW_WR_IMMDLEN_V(mpalen));
931 req->flowid_len16 = cpu_to_be32(
932 FW_WR_FLOWID_V(ep->hwtid) |
933 FW_WR_LEN16_V(wrlen >> 4));
934 req->plen = cpu_to_be32(mpalen);
935 req->tunnel_to_proxy = cpu_to_be32(
936 FW_OFLD_TX_DATA_WR_FLUSH_F |
937 FW_OFLD_TX_DATA_WR_SHOVE_F);
939 mpa = (struct mpa_message *)(req + 1);
940 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
944 mpa->flags |= MPA_CRC;
945 if (markers_enabled) {
946 mpa->flags |= MPA_MARKERS;
947 ep->mpa_attr.recv_marker_enabled = 1;
949 ep->mpa_attr.recv_marker_enabled = 0;
951 if (mpa_rev_to_use == 2)
952 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
954 mpa->private_data_size = htons(ep->plen);
955 mpa->revision = mpa_rev_to_use;
956 if (mpa_rev_to_use == 1) {
957 ep->tried_with_mpa_v1 = 1;
958 ep->retry_with_mpa_v1 = 0;
961 if (mpa_rev_to_use == 2) {
962 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
963 sizeof (struct mpa_v2_conn_params));
964 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
966 mpa_v2_params.ird = htons((u16)ep->ird);
967 mpa_v2_params.ord = htons((u16)ep->ord);
970 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
971 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
973 htons(MPA_V2_RDMA_WRITE_RTR);
974 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
976 htons(MPA_V2_RDMA_READ_RTR);
978 memcpy(mpa->private_data, &mpa_v2_params,
979 sizeof(struct mpa_v2_conn_params));
982 memcpy(mpa->private_data +
983 sizeof(struct mpa_v2_conn_params),
984 ep->mpa_pkt + sizeof(*mpa), ep->plen);
987 memcpy(mpa->private_data,
988 ep->mpa_pkt + sizeof(*mpa), ep->plen);
991 * Reference the mpa skb. This ensures the data area
992 * will remain in memory until the hw acks the tx.
993 * Function fw4_ack() will deref it.
996 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
999 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1003 __state_set(&ep->com, MPA_REQ_SENT);
1004 ep->mpa_attr.initiator = 1;
1005 ep->snd_seq += mpalen;
1009 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1012 struct fw_ofld_tx_data_wr *req;
1013 struct mpa_message *mpa;
1014 struct sk_buff *skb;
1015 struct mpa_v2_conn_params mpa_v2_params;
1017 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1019 mpalen = sizeof(*mpa) + plen;
1020 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1021 mpalen += sizeof(struct mpa_v2_conn_params);
1022 wrlen = roundup(mpalen + sizeof *req, 16);
1024 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1026 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1029 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1031 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1032 memset(req, 0, wrlen);
1033 req->op_to_immdlen = cpu_to_be32(
1034 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1036 FW_WR_IMMDLEN_V(mpalen));
1037 req->flowid_len16 = cpu_to_be32(
1038 FW_WR_FLOWID_V(ep->hwtid) |
1039 FW_WR_LEN16_V(wrlen >> 4));
1040 req->plen = cpu_to_be32(mpalen);
1041 req->tunnel_to_proxy = cpu_to_be32(
1042 FW_OFLD_TX_DATA_WR_FLUSH_F |
1043 FW_OFLD_TX_DATA_WR_SHOVE_F);
1045 mpa = (struct mpa_message *)(req + 1);
1046 memset(mpa, 0, sizeof(*mpa));
1047 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1048 mpa->flags = MPA_REJECT;
1049 mpa->revision = ep->mpa_attr.version;
1050 mpa->private_data_size = htons(plen);
1052 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1053 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1054 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1055 sizeof (struct mpa_v2_conn_params));
1056 mpa_v2_params.ird = htons(((u16)ep->ird) |
1057 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1059 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1061 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1062 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1063 FW_RI_INIT_P2PTYPE_READ_REQ ?
1064 MPA_V2_RDMA_READ_RTR : 0) : 0));
1065 memcpy(mpa->private_data, &mpa_v2_params,
1066 sizeof(struct mpa_v2_conn_params));
1069 memcpy(mpa->private_data +
1070 sizeof(struct mpa_v2_conn_params), pdata, plen);
1073 memcpy(mpa->private_data, pdata, plen);
1076 * Reference the mpa skb again. This ensures the data area
1077 * will remain in memory until the hw acks the tx.
1078 * Function fw4_ack() will deref it.
1081 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1082 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1083 BUG_ON(ep->mpa_skb);
1085 ep->snd_seq += mpalen;
1086 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1089 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1092 struct fw_ofld_tx_data_wr *req;
1093 struct mpa_message *mpa;
1094 struct sk_buff *skb;
1095 struct mpa_v2_conn_params mpa_v2_params;
1097 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1099 mpalen = sizeof(*mpa) + plen;
1100 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1101 mpalen += sizeof(struct mpa_v2_conn_params);
1102 wrlen = roundup(mpalen + sizeof *req, 16);
1104 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1106 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1109 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1111 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1112 memset(req, 0, wrlen);
1113 req->op_to_immdlen = cpu_to_be32(
1114 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1116 FW_WR_IMMDLEN_V(mpalen));
1117 req->flowid_len16 = cpu_to_be32(
1118 FW_WR_FLOWID_V(ep->hwtid) |
1119 FW_WR_LEN16_V(wrlen >> 4));
1120 req->plen = cpu_to_be32(mpalen);
1121 req->tunnel_to_proxy = cpu_to_be32(
1122 FW_OFLD_TX_DATA_WR_FLUSH_F |
1123 FW_OFLD_TX_DATA_WR_SHOVE_F);
1125 mpa = (struct mpa_message *)(req + 1);
1126 memset(mpa, 0, sizeof(*mpa));
1127 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1129 if (ep->mpa_attr.crc_enabled)
1130 mpa->flags |= MPA_CRC;
1131 if (ep->mpa_attr.recv_marker_enabled)
1132 mpa->flags |= MPA_MARKERS;
1133 mpa->revision = ep->mpa_attr.version;
1134 mpa->private_data_size = htons(plen);
1136 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1137 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1138 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1139 sizeof (struct mpa_v2_conn_params));
1140 mpa_v2_params.ird = htons((u16)ep->ird);
1141 mpa_v2_params.ord = htons((u16)ep->ord);
1142 if (peer2peer && (ep->mpa_attr.p2p_type !=
1143 FW_RI_INIT_P2PTYPE_DISABLED)) {
1144 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1146 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1147 mpa_v2_params.ord |=
1148 htons(MPA_V2_RDMA_WRITE_RTR);
1149 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1150 mpa_v2_params.ord |=
1151 htons(MPA_V2_RDMA_READ_RTR);
1154 memcpy(mpa->private_data, &mpa_v2_params,
1155 sizeof(struct mpa_v2_conn_params));
1158 memcpy(mpa->private_data +
1159 sizeof(struct mpa_v2_conn_params), pdata, plen);
1162 memcpy(mpa->private_data, pdata, plen);
1165 * Reference the mpa skb. This ensures the data area
1166 * will remain in memory until the hw acks the tx.
1167 * Function fw4_ack() will deref it.
1170 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1172 __state_set(&ep->com, MPA_REP_SENT);
1173 ep->snd_seq += mpalen;
1174 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1177 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1180 struct cpl_act_establish *req = cplhdr(skb);
1181 unsigned int tid = GET_TID(req);
1182 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1183 struct tid_info *t = dev->rdev.lldi.tids;
1186 ep = lookup_atid(t, atid);
1188 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1189 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1191 mutex_lock(&ep->com.mutex);
1192 dst_confirm(ep->dst);
1194 /* setup the hwtid for this connection */
1196 cxgb4_insert_tid(t, ep, tid);
1199 ep->snd_seq = be32_to_cpu(req->snd_isn);
1200 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1202 set_emss(ep, ntohs(req->tcp_opt));
1204 /* dealloc the atid */
1205 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1206 cxgb4_free_atid(t, atid);
1207 set_bit(ACT_ESTAB, &ep->com.history);
1209 /* start MPA negotiation */
1210 ret = send_flowc(ep);
1213 if (ep->retry_with_mpa_v1)
1214 ret = send_mpa_req(ep, skb, 1);
1216 ret = send_mpa_req(ep, skb, mpa_rev);
1219 mutex_unlock(&ep->com.mutex);
1222 mutex_unlock(&ep->com.mutex);
1223 connect_reply_upcall(ep, -ENOMEM);
1224 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1228 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1230 struct iw_cm_event event;
1232 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1233 memset(&event, 0, sizeof(event));
1234 event.event = IW_CM_EVENT_CLOSE;
1235 event.status = status;
1236 if (ep->com.cm_id) {
1237 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1238 ep, ep->com.cm_id, ep->hwtid);
1239 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1240 deref_cm_id(&ep->com);
1241 set_bit(CLOSE_UPCALL, &ep->com.history);
1245 static void peer_close_upcall(struct c4iw_ep *ep)
1247 struct iw_cm_event event;
1249 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1250 memset(&event, 0, sizeof(event));
1251 event.event = IW_CM_EVENT_DISCONNECT;
1252 if (ep->com.cm_id) {
1253 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1254 ep, ep->com.cm_id, ep->hwtid);
1255 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1256 set_bit(DISCONN_UPCALL, &ep->com.history);
1260 static void peer_abort_upcall(struct c4iw_ep *ep)
1262 struct iw_cm_event event;
1264 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1265 memset(&event, 0, sizeof(event));
1266 event.event = IW_CM_EVENT_CLOSE;
1267 event.status = -ECONNRESET;
1268 if (ep->com.cm_id) {
1269 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1270 ep->com.cm_id, ep->hwtid);
1271 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1272 deref_cm_id(&ep->com);
1273 set_bit(ABORT_UPCALL, &ep->com.history);
1277 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1279 struct iw_cm_event event;
1281 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1282 memset(&event, 0, sizeof(event));
1283 event.event = IW_CM_EVENT_CONNECT_REPLY;
1284 event.status = status;
1285 memcpy(&event.local_addr, &ep->com.local_addr,
1286 sizeof(ep->com.local_addr));
1287 memcpy(&event.remote_addr, &ep->com.remote_addr,
1288 sizeof(ep->com.remote_addr));
1290 if ((status == 0) || (status == -ECONNREFUSED)) {
1291 if (!ep->tried_with_mpa_v1) {
1292 /* this means MPA_v2 is used */
1293 event.ord = ep->ird;
1294 event.ird = ep->ord;
1295 event.private_data_len = ep->plen -
1296 sizeof(struct mpa_v2_conn_params);
1297 event.private_data = ep->mpa_pkt +
1298 sizeof(struct mpa_message) +
1299 sizeof(struct mpa_v2_conn_params);
1301 /* this means MPA_v1 is used */
1302 event.ord = cur_max_read_depth(ep->com.dev);
1303 event.ird = cur_max_read_depth(ep->com.dev);
1304 event.private_data_len = ep->plen;
1305 event.private_data = ep->mpa_pkt +
1306 sizeof(struct mpa_message);
1310 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1312 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1313 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1316 deref_cm_id(&ep->com);
1319 static int connect_request_upcall(struct c4iw_ep *ep)
1321 struct iw_cm_event event;
1324 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1325 memset(&event, 0, sizeof(event));
1326 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1327 memcpy(&event.local_addr, &ep->com.local_addr,
1328 sizeof(ep->com.local_addr));
1329 memcpy(&event.remote_addr, &ep->com.remote_addr,
1330 sizeof(ep->com.remote_addr));
1331 event.provider_data = ep;
1332 if (!ep->tried_with_mpa_v1) {
1333 /* this means MPA_v2 is used */
1334 event.ord = ep->ord;
1335 event.ird = ep->ird;
1336 event.private_data_len = ep->plen -
1337 sizeof(struct mpa_v2_conn_params);
1338 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1339 sizeof(struct mpa_v2_conn_params);
1341 /* this means MPA_v1 is used. Send max supported */
1342 event.ord = cur_max_read_depth(ep->com.dev);
1343 event.ird = cur_max_read_depth(ep->com.dev);
1344 event.private_data_len = ep->plen;
1345 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1347 c4iw_get_ep(&ep->com);
1348 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1351 c4iw_put_ep(&ep->com);
1352 set_bit(CONNREQ_UPCALL, &ep->com.history);
1353 c4iw_put_ep(&ep->parent_ep->com);
1357 static void established_upcall(struct c4iw_ep *ep)
1359 struct iw_cm_event event;
1361 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1362 memset(&event, 0, sizeof(event));
1363 event.event = IW_CM_EVENT_ESTABLISHED;
1364 event.ird = ep->ord;
1365 event.ord = ep->ird;
1366 if (ep->com.cm_id) {
1367 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1368 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1369 set_bit(ESTAB_UPCALL, &ep->com.history);
1373 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1375 struct sk_buff *skb;
1376 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1379 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1380 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1382 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1387 * If we couldn't specify the entire rcv window at connection setup
1388 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1389 * then add the overage in to the credits returned.
1391 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1392 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1394 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1395 RX_DACK_MODE_V(dack_mode);
1397 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1400 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1404 #define RELAXED_IRD_NEGOTIATION 1
1407 * process_mpa_reply - process streaming mode MPA reply
1411 * 0 upon success indicating a connect request was delivered to the ULP
1412 * or the mpa request is incomplete but valid so far.
1414 * 1 if a failure requires the caller to close the connection.
1416 * 2 if a failure requires the caller to abort the connection.
1418 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1420 struct mpa_message *mpa;
1421 struct mpa_v2_conn_params *mpa_v2_params;
1423 u16 resp_ird, resp_ord;
1424 u8 rtr_mismatch = 0, insuff_ird = 0;
1425 struct c4iw_qp_attributes attrs;
1426 enum c4iw_qp_attr_mask mask;
1430 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1433 * If we get more than the supported amount of private data
1434 * then we must fail this connection.
1436 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1438 goto err_stop_timer;
1442 * copy the new data into our accumulation buffer.
1444 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1446 ep->mpa_pkt_len += skb->len;
1449 * if we don't even have the mpa message, then bail.
1451 if (ep->mpa_pkt_len < sizeof(*mpa))
1453 mpa = (struct mpa_message *) ep->mpa_pkt;
1455 /* Validate MPA header. */
1456 if (mpa->revision > mpa_rev) {
1457 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1458 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1460 goto err_stop_timer;
1462 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1464 goto err_stop_timer;
1467 plen = ntohs(mpa->private_data_size);
1470 * Fail if there's too much private data.
1472 if (plen > MPA_MAX_PRIVATE_DATA) {
1474 goto err_stop_timer;
1478 * If plen does not account for pkt size
1480 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1482 goto err_stop_timer;
1485 ep->plen = (u8) plen;
1488 * If we don't have all the pdata yet, then bail.
1489 * We'll continue process when more data arrives.
1491 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1494 if (mpa->flags & MPA_REJECT) {
1495 err = -ECONNREFUSED;
1496 goto err_stop_timer;
1500 * Stop mpa timer. If it expired, then
1501 * we ignore the MPA reply. process_timeout()
1502 * will abort the connection.
1504 if (stop_ep_timer(ep))
1508 * If we get here we have accumulated the entire mpa
1509 * start reply message including private data. And
1510 * the MPA header is valid.
1512 __state_set(&ep->com, FPDU_MODE);
1513 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1514 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1515 ep->mpa_attr.version = mpa->revision;
1516 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1518 if (mpa->revision == 2) {
1519 ep->mpa_attr.enhanced_rdma_conn =
1520 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1521 if (ep->mpa_attr.enhanced_rdma_conn) {
1522 mpa_v2_params = (struct mpa_v2_conn_params *)
1523 (ep->mpa_pkt + sizeof(*mpa));
1524 resp_ird = ntohs(mpa_v2_params->ird) &
1525 MPA_V2_IRD_ORD_MASK;
1526 resp_ord = ntohs(mpa_v2_params->ord) &
1527 MPA_V2_IRD_ORD_MASK;
1528 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1529 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1532 * This is a double-check. Ideally, below checks are
1533 * not required since ird/ord stuff has been taken
1534 * care of in c4iw_accept_cr
1536 if (ep->ird < resp_ord) {
1537 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1538 ep->com.dev->rdev.lldi.max_ordird_qp)
1542 } else if (ep->ird > resp_ord) {
1545 if (ep->ord > resp_ird) {
1546 if (RELAXED_IRD_NEGOTIATION)
1557 if (ntohs(mpa_v2_params->ird) &
1558 MPA_V2_PEER2PEER_MODEL) {
1559 if (ntohs(mpa_v2_params->ord) &
1560 MPA_V2_RDMA_WRITE_RTR)
1561 ep->mpa_attr.p2p_type =
1562 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1563 else if (ntohs(mpa_v2_params->ord) &
1564 MPA_V2_RDMA_READ_RTR)
1565 ep->mpa_attr.p2p_type =
1566 FW_RI_INIT_P2PTYPE_READ_REQ;
1569 } else if (mpa->revision == 1)
1571 ep->mpa_attr.p2p_type = p2p_type;
1573 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1574 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1575 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1576 ep->mpa_attr.recv_marker_enabled,
1577 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1578 ep->mpa_attr.p2p_type, p2p_type);
1581 * If responder's RTR does not match with that of initiator, assign
1582 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1583 * generated when moving QP to RTS state.
1584 * A TERM message will be sent after QP has moved to RTS state
1586 if ((ep->mpa_attr.version == 2) && peer2peer &&
1587 (ep->mpa_attr.p2p_type != p2p_type)) {
1588 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1592 attrs.mpa_attr = ep->mpa_attr;
1593 attrs.max_ird = ep->ird;
1594 attrs.max_ord = ep->ord;
1595 attrs.llp_stream_handle = ep;
1596 attrs.next_state = C4IW_QP_STATE_RTS;
1598 mask = C4IW_QP_ATTR_NEXT_STATE |
1599 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1600 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1602 /* bind QP and TID with INIT_WR */
1603 err = c4iw_modify_qp(ep->com.qp->rhp,
1604 ep->com.qp, mask, &attrs, 1);
1609 * If responder's RTR requirement did not match with what initiator
1610 * supports, generate TERM message
1613 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1614 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1615 attrs.ecode = MPA_NOMATCH_RTR;
1616 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1617 attrs.send_term = 1;
1618 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1619 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1626 * Generate TERM if initiator IRD is not sufficient for responder
1627 * provided ORD. Currently, we do the same behaviour even when
1628 * responder provided IRD is also not sufficient as regards to
1632 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1634 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1635 attrs.ecode = MPA_INSUFF_IRD;
1636 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1637 attrs.send_term = 1;
1638 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1639 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1650 connect_reply_upcall(ep, err);
1655 * process_mpa_request - process streaming mode MPA request
1659 * 0 upon success indicating a connect request was delivered to the ULP
1660 * or the mpa request is incomplete but valid so far.
1662 * 1 if a failure requires the caller to close the connection.
1664 * 2 if a failure requires the caller to abort the connection.
1666 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1668 struct mpa_message *mpa;
1669 struct mpa_v2_conn_params *mpa_v2_params;
1672 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1675 * If we get more than the supported amount of private data
1676 * then we must fail this connection.
1678 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1679 goto err_stop_timer;
1681 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1684 * Copy the new data into our accumulation buffer.
1686 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1688 ep->mpa_pkt_len += skb->len;
1691 * If we don't even have the mpa message, then bail.
1692 * We'll continue process when more data arrives.
1694 if (ep->mpa_pkt_len < sizeof(*mpa))
1697 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1698 mpa = (struct mpa_message *) ep->mpa_pkt;
1701 * Validate MPA Header.
1703 if (mpa->revision > mpa_rev) {
1704 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1705 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1706 goto err_stop_timer;
1709 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1710 goto err_stop_timer;
1712 plen = ntohs(mpa->private_data_size);
1715 * Fail if there's too much private data.
1717 if (plen > MPA_MAX_PRIVATE_DATA)
1718 goto err_stop_timer;
1721 * If plen does not account for pkt size
1723 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1724 goto err_stop_timer;
1725 ep->plen = (u8) plen;
1728 * If we don't have all the pdata yet, then bail.
1730 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1734 * If we get here we have accumulated the entire mpa
1735 * start reply message including private data.
1737 ep->mpa_attr.initiator = 0;
1738 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1739 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1740 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1741 ep->mpa_attr.version = mpa->revision;
1742 if (mpa->revision == 1)
1743 ep->tried_with_mpa_v1 = 1;
1744 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1746 if (mpa->revision == 2) {
1747 ep->mpa_attr.enhanced_rdma_conn =
1748 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1749 if (ep->mpa_attr.enhanced_rdma_conn) {
1750 mpa_v2_params = (struct mpa_v2_conn_params *)
1751 (ep->mpa_pkt + sizeof(*mpa));
1752 ep->ird = ntohs(mpa_v2_params->ird) &
1753 MPA_V2_IRD_ORD_MASK;
1754 ep->ird = min_t(u32, ep->ird,
1755 cur_max_read_depth(ep->com.dev));
1756 ep->ord = ntohs(mpa_v2_params->ord) &
1757 MPA_V2_IRD_ORD_MASK;
1758 ep->ord = min_t(u32, ep->ord,
1759 cur_max_read_depth(ep->com.dev));
1760 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1762 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1764 if (ntohs(mpa_v2_params->ord) &
1765 MPA_V2_RDMA_WRITE_RTR)
1766 ep->mpa_attr.p2p_type =
1767 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1768 else if (ntohs(mpa_v2_params->ord) &
1769 MPA_V2_RDMA_READ_RTR)
1770 ep->mpa_attr.p2p_type =
1771 FW_RI_INIT_P2PTYPE_READ_REQ;
1774 } else if (mpa->revision == 1)
1776 ep->mpa_attr.p2p_type = p2p_type;
1778 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1779 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1780 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1781 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1782 ep->mpa_attr.p2p_type);
1784 __state_set(&ep->com, MPA_REQ_RCVD);
1787 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1788 if (ep->parent_ep->com.state != DEAD) {
1789 if (connect_request_upcall(ep))
1790 goto err_unlock_parent;
1792 goto err_unlock_parent;
1794 mutex_unlock(&ep->parent_ep->com.mutex);
1798 mutex_unlock(&ep->parent_ep->com.mutex);
1801 (void)stop_ep_timer(ep);
1806 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1809 struct cpl_rx_data *hdr = cplhdr(skb);
1810 unsigned int dlen = ntohs(hdr->len);
1811 unsigned int tid = GET_TID(hdr);
1812 __u8 status = hdr->status;
1815 ep = get_ep_from_tid(dev, tid);
1818 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1819 skb_pull(skb, sizeof(*hdr));
1820 skb_trim(skb, dlen);
1821 mutex_lock(&ep->com.mutex);
1823 switch (ep->com.state) {
1825 update_rx_credits(ep, dlen);
1826 ep->rcv_seq += dlen;
1827 disconnect = process_mpa_reply(ep, skb);
1830 update_rx_credits(ep, dlen);
1831 ep->rcv_seq += dlen;
1832 disconnect = process_mpa_request(ep, skb);
1835 struct c4iw_qp_attributes attrs;
1837 update_rx_credits(ep, dlen);
1838 BUG_ON(!ep->com.qp);
1840 pr_err("%s Unexpected streaming data." \
1841 " qpid %u ep %p state %d tid %u status %d\n",
1842 __func__, ep->com.qp->wq.sq.qid, ep,
1843 ep->com.state, ep->hwtid, status);
1844 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1845 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1846 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1853 mutex_unlock(&ep->com.mutex);
1855 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1856 c4iw_put_ep(&ep->com);
1860 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1863 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1865 unsigned int tid = GET_TID(rpl);
1867 ep = get_ep_from_tid(dev, tid);
1869 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1872 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1873 mutex_lock(&ep->com.mutex);
1874 switch (ep->com.state) {
1876 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1877 __state_set(&ep->com, DEAD);
1881 printk(KERN_ERR "%s ep %p state %d\n",
1882 __func__, ep, ep->com.state);
1885 mutex_unlock(&ep->com.mutex);
1888 release_ep_resources(ep);
1889 c4iw_put_ep(&ep->com);
1893 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1895 struct sk_buff *skb;
1896 struct fw_ofld_connection_wr *req;
1897 unsigned int mtu_idx;
1899 struct sockaddr_in *sin;
1902 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1903 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1904 memset(req, 0, sizeof(*req));
1905 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1906 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1907 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1908 ep->com.dev->rdev.lldi.ports[0],
1910 sin = (struct sockaddr_in *)&ep->com.local_addr;
1911 req->le.lport = sin->sin_port;
1912 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1913 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1914 req->le.pport = sin->sin_port;
1915 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1916 req->tcb.t_state_to_astid =
1917 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1918 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1919 req->tcb.cplrxdataack_cplpassacceptrpl =
1920 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1921 req->tcb.tx_max = (__force __be32) jiffies;
1922 req->tcb.rcv_adv = htons(1);
1923 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1924 enable_tcp_timestamps,
1925 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1926 wscale = cxgb_compute_wscale(rcv_win);
1929 * Specify the largest window that will fit in opt0. The
1930 * remainder will be specified in the rx_data_ack.
1932 win = ep->rcv_win >> 10;
1933 if (win > RCV_BUFSIZ_M)
1936 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1937 (nocong ? NO_CONG_F : 0) |
1940 WND_SCALE_V(wscale) |
1941 MSS_IDX_V(mtu_idx) |
1942 L2T_IDX_V(ep->l2t->idx) |
1943 TX_CHAN_V(ep->tx_chan) |
1944 SMAC_SEL_V(ep->smac_idx) |
1945 DSCP_V(ep->tos >> 2) |
1946 ULP_MODE_V(ULP_MODE_TCPDDP) |
1948 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1949 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1951 CCTRL_ECN_V(enable_ecn) |
1952 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1953 if (enable_tcp_timestamps)
1954 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1955 if (enable_tcp_sack)
1956 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1957 if (wscale && enable_tcp_window_scaling)
1958 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1959 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1960 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1961 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1962 set_bit(ACT_OFLD_CONN, &ep->com.history);
1963 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1967 * Some of the error codes above implicitly indicate that there is no TID
1968 * allocated with the result of an ACT_OPEN. We use this predicate to make
1971 static inline int act_open_has_tid(int status)
1973 return (status != CPL_ERR_TCAM_PARITY &&
1974 status != CPL_ERR_TCAM_MISS &&
1975 status != CPL_ERR_TCAM_FULL &&
1976 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1977 status != CPL_ERR_CONN_EXIST);
1980 static char *neg_adv_str(unsigned int status)
1983 case CPL_ERR_RTX_NEG_ADVICE:
1984 return "Retransmit timeout";
1985 case CPL_ERR_PERSIST_NEG_ADVICE:
1986 return "Persist timeout";
1987 case CPL_ERR_KEEPALV_NEG_ADVICE:
1988 return "Keepalive timeout";
1994 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1996 ep->snd_win = snd_win;
1997 ep->rcv_win = rcv_win;
1998 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
2001 #define ACT_OPEN_RETRY_COUNT 2
2003 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2004 struct dst_entry *dst, struct c4iw_dev *cdev,
2005 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2007 struct neighbour *n;
2009 struct net_device *pdev;
2011 n = dst_neigh_lookup(dst, peer_ip);
2017 if (n->dev->flags & IFF_LOOPBACK) {
2019 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2020 else if (IS_ENABLED(CONFIG_IPV6))
2021 for_each_netdev(&init_net, pdev) {
2022 if (ipv6_chk_addr(&init_net,
2023 (struct in6_addr *)peer_ip,
2034 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2035 n, pdev, rt_tos2priority(tos));
2040 ep->mtu = pdev->mtu;
2041 ep->tx_chan = cxgb4_port_chan(pdev);
2042 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2043 cxgb4_port_viid(pdev));
2044 step = cdev->rdev.lldi.ntxq /
2045 cdev->rdev.lldi.nchan;
2046 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2047 step = cdev->rdev.lldi.nrxq /
2048 cdev->rdev.lldi.nchan;
2049 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2050 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2051 cxgb4_port_idx(pdev) * step];
2052 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2055 pdev = get_real_dev(n->dev);
2056 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2060 ep->mtu = dst_mtu(dst);
2061 ep->tx_chan = cxgb4_port_chan(pdev);
2062 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2063 cxgb4_port_viid(pdev));
2064 step = cdev->rdev.lldi.ntxq /
2065 cdev->rdev.lldi.nchan;
2066 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2067 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2068 step = cdev->rdev.lldi.nrxq /
2069 cdev->rdev.lldi.nchan;
2070 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2071 cxgb4_port_idx(pdev) * step];
2072 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2075 ep->retry_with_mpa_v1 = 0;
2076 ep->tried_with_mpa_v1 = 0;
2088 static int c4iw_reconnect(struct c4iw_ep *ep)
2092 struct sockaddr_in *laddr = (struct sockaddr_in *)
2093 &ep->com.cm_id->m_local_addr;
2094 struct sockaddr_in *raddr = (struct sockaddr_in *)
2095 &ep->com.cm_id->m_remote_addr;
2096 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2097 &ep->com.cm_id->m_local_addr;
2098 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2099 &ep->com.cm_id->m_remote_addr;
2103 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2104 init_timer(&ep->timer);
2105 c4iw_init_wr_wait(&ep->com.wr_wait);
2107 /* When MPA revision is different on nodes, the node with MPA_rev=2
2108 * tries to reconnect with MPA_rev 1 for the same EP through
2109 * c4iw_reconnect(), where the same EP is assigned with new tid for
2110 * further connection establishment. As we are using the same EP pointer
2111 * for reconnect, few skbs are used during the previous c4iw_connect(),
2112 * which leaves the EP with inadequate skbs for further
2113 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2114 * skb_list() during peer_abort(). Allocate skbs which is already used.
2116 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2117 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2123 * Allocate an active TID to initiate a TCP connection.
2125 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2126 if (ep->atid == -1) {
2127 pr_err("%s - cannot alloc atid.\n", __func__);
2131 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2134 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2135 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2136 laddr->sin_addr.s_addr,
2137 raddr->sin_addr.s_addr,
2139 raddr->sin_port, ep->com.cm_id->tos);
2141 ra = (__u8 *)&raddr->sin_addr;
2143 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2145 laddr6->sin6_addr.s6_addr,
2146 raddr6->sin6_addr.s6_addr,
2148 raddr6->sin6_port, 0,
2149 raddr6->sin6_scope_id);
2151 ra = (__u8 *)&raddr6->sin6_addr;
2154 pr_err("%s - cannot find route.\n", __func__);
2155 err = -EHOSTUNREACH;
2158 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2159 ep->com.dev->rdev.lldi.adapter_type,
2160 ep->com.cm_id->tos);
2162 pr_err("%s - cannot alloc l2e.\n", __func__);
2166 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2167 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2170 state_set(&ep->com, CONNECTING);
2171 ep->tos = ep->com.cm_id->tos;
2173 /* send connect request to rnic */
2174 err = send_connect(ep);
2178 cxgb4_l2t_release(ep->l2t);
2180 dst_release(ep->dst);
2182 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2183 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2186 * remember to send notification to upper layer.
2187 * We are in here so the upper layer is not aware that this is
2188 * re-connect attempt and so, upper layer is still waiting for
2189 * response of 1st connect request.
2191 connect_reply_upcall(ep, -ECONNRESET);
2193 c4iw_put_ep(&ep->com);
2198 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2201 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2202 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2203 ntohl(rpl->atid_status)));
2204 struct tid_info *t = dev->rdev.lldi.tids;
2205 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2206 struct sockaddr_in *la;
2207 struct sockaddr_in *ra;
2208 struct sockaddr_in6 *la6;
2209 struct sockaddr_in6 *ra6;
2212 ep = lookup_atid(t, atid);
2213 la = (struct sockaddr_in *)&ep->com.local_addr;
2214 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2215 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2216 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2218 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2219 status, status2errno(status));
2221 if (cxgb_is_neg_adv(status)) {
2222 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2223 __func__, atid, status, neg_adv_str(status));
2224 ep->stats.connect_neg_adv++;
2225 mutex_lock(&dev->rdev.stats.lock);
2226 dev->rdev.stats.neg_adv++;
2227 mutex_unlock(&dev->rdev.stats.lock);
2231 set_bit(ACT_OPEN_RPL, &ep->com.history);
2234 * Log interesting failures.
2237 case CPL_ERR_CONN_RESET:
2238 case CPL_ERR_CONN_TIMEDOUT:
2240 case CPL_ERR_TCAM_FULL:
2241 mutex_lock(&dev->rdev.stats.lock);
2242 dev->rdev.stats.tcam_full++;
2243 mutex_unlock(&dev->rdev.stats.lock);
2244 if (ep->com.local_addr.ss_family == AF_INET &&
2245 dev->rdev.lldi.enable_fw_ofld_conn) {
2246 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2247 ntohl(rpl->atid_status))));
2253 case CPL_ERR_CONN_EXIST:
2254 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2255 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2256 if (ep->com.remote_addr.ss_family == AF_INET6) {
2257 struct sockaddr_in6 *sin6 =
2258 (struct sockaddr_in6 *)
2259 &ep->com.local_addr;
2261 ep->com.dev->rdev.lldi.ports[0],
2263 &sin6->sin6_addr.s6_addr, 1);
2265 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2267 cxgb4_free_atid(t, atid);
2268 dst_release(ep->dst);
2269 cxgb4_l2t_release(ep->l2t);
2275 if (ep->com.local_addr.ss_family == AF_INET) {
2276 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2277 atid, status, status2errno(status),
2278 &la->sin_addr.s_addr, ntohs(la->sin_port),
2279 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2281 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2282 atid, status, status2errno(status),
2283 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2284 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2290 connect_reply_upcall(ep, status2errno(status));
2291 state_set(&ep->com, DEAD);
2293 if (ep->com.remote_addr.ss_family == AF_INET6) {
2294 struct sockaddr_in6 *sin6 =
2295 (struct sockaddr_in6 *)&ep->com.local_addr;
2296 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2297 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2299 if (status && act_open_has_tid(status))
2300 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2302 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2303 cxgb4_free_atid(t, atid);
2304 dst_release(ep->dst);
2305 cxgb4_l2t_release(ep->l2t);
2306 c4iw_put_ep(&ep->com);
2311 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2313 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2314 unsigned int stid = GET_TID(rpl);
2315 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2318 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2321 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2322 rpl->status, status2errno(rpl->status));
2323 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2324 c4iw_put_ep(&ep->com);
2329 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2331 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2332 unsigned int stid = GET_TID(rpl);
2333 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2335 PDBG("%s ep %p\n", __func__, ep);
2336 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2337 c4iw_put_ep(&ep->com);
2341 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2342 struct cpl_pass_accept_req *req)
2344 struct cpl_pass_accept_rpl *rpl;
2345 unsigned int mtu_idx;
2349 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2351 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2353 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2354 BUG_ON(skb_cloned(skb));
2358 if (!is_t4(adapter_type)) {
2359 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2361 INIT_TP_WR(rpl5, ep->hwtid);
2363 skb_trim(skb, sizeof(*rpl));
2364 INIT_TP_WR(rpl, ep->hwtid);
2366 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2369 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2370 enable_tcp_timestamps && req->tcpopt.tstamp,
2371 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2372 wscale = cxgb_compute_wscale(rcv_win);
2375 * Specify the largest window that will fit in opt0. The
2376 * remainder will be specified in the rx_data_ack.
2378 win = ep->rcv_win >> 10;
2379 if (win > RCV_BUFSIZ_M)
2381 opt0 = (nocong ? NO_CONG_F : 0) |
2384 WND_SCALE_V(wscale) |
2385 MSS_IDX_V(mtu_idx) |
2386 L2T_IDX_V(ep->l2t->idx) |
2387 TX_CHAN_V(ep->tx_chan) |
2388 SMAC_SEL_V(ep->smac_idx) |
2389 DSCP_V(ep->tos >> 2) |
2390 ULP_MODE_V(ULP_MODE_TCPDDP) |
2392 opt2 = RX_CHANNEL_V(0) |
2393 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2395 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2396 opt2 |= TSTAMPS_EN_F;
2397 if (enable_tcp_sack && req->tcpopt.sack)
2399 if (wscale && enable_tcp_window_scaling)
2400 opt2 |= WND_SCALE_EN_F;
2402 const struct tcphdr *tcph;
2403 u32 hlen = ntohl(req->hdr_len);
2405 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2406 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2409 tcph = (const void *)(req + 1) +
2410 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2411 if (tcph->ece && tcph->cwr)
2412 opt2 |= CCTRL_ECN_V(1);
2414 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2415 u32 isn = (prandom_u32() & ~7UL) - 1;
2416 opt2 |= T5_OPT_2_VALID_F;
2417 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2420 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2423 rpl5->iss = cpu_to_be32(isn);
2424 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2427 rpl->opt0 = cpu_to_be64(opt0);
2428 rpl->opt2 = cpu_to_be32(opt2);
2429 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2430 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2432 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2435 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2437 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2438 BUG_ON(skb_cloned(skb));
2439 skb_trim(skb, sizeof(struct cpl_tid_release));
2440 release_tid(&dev->rdev, hwtid, skb);
2444 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2446 struct c4iw_ep *child_ep = NULL, *parent_ep;
2447 struct cpl_pass_accept_req *req = cplhdr(skb);
2448 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2449 struct tid_info *t = dev->rdev.lldi.tids;
2450 unsigned int hwtid = GET_TID(req);
2451 struct dst_entry *dst;
2452 __u8 local_ip[16], peer_ip[16];
2453 __be16 local_port, peer_port;
2454 struct sockaddr_in6 *sin6;
2456 u16 peer_mss = ntohs(req->tcpopt.mss);
2458 unsigned short hdrs;
2459 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2461 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2463 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2467 if (state_read(&parent_ep->com) != LISTEN) {
2468 PDBG("%s - listening ep not in LISTEN\n", __func__);
2472 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2473 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2475 /* Find output route */
2477 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2478 , __func__, parent_ep, hwtid,
2479 local_ip, peer_ip, ntohs(local_port),
2480 ntohs(peer_port), peer_mss);
2481 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2482 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2483 local_port, peer_port, tos);
2485 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2486 , __func__, parent_ep, hwtid,
2487 local_ip, peer_ip, ntohs(local_port),
2488 ntohs(peer_port), peer_mss);
2489 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2490 local_ip, peer_ip, local_port, peer_port,
2491 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2492 ((struct sockaddr_in6 *)
2493 &parent_ep->com.local_addr)->sin6_scope_id);
2496 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2501 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2503 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2509 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2510 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2512 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2519 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2520 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2521 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2522 child_ep->mtu = peer_mss + hdrs;
2524 skb_queue_head_init(&child_ep->com.ep_skb_list);
2525 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2528 state_set(&child_ep->com, CONNECTING);
2529 child_ep->com.dev = dev;
2530 child_ep->com.cm_id = NULL;
2533 struct sockaddr_in *sin = (struct sockaddr_in *)
2534 &child_ep->com.local_addr;
2536 sin->sin_family = AF_INET;
2537 sin->sin_port = local_port;
2538 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2540 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2541 sin->sin_family = AF_INET;
2542 sin->sin_port = ((struct sockaddr_in *)
2543 &parent_ep->com.local_addr)->sin_port;
2544 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2546 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2547 sin->sin_family = AF_INET;
2548 sin->sin_port = peer_port;
2549 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2551 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2552 sin6->sin6_family = PF_INET6;
2553 sin6->sin6_port = local_port;
2554 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2556 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2557 sin6->sin6_family = PF_INET6;
2558 sin6->sin6_port = ((struct sockaddr_in6 *)
2559 &parent_ep->com.local_addr)->sin6_port;
2560 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2562 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2563 sin6->sin6_family = PF_INET6;
2564 sin6->sin6_port = peer_port;
2565 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2568 c4iw_get_ep(&parent_ep->com);
2569 child_ep->parent_ep = parent_ep;
2570 child_ep->tos = tos;
2571 child_ep->dst = dst;
2572 child_ep->hwtid = hwtid;
2574 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2575 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2577 init_timer(&child_ep->timer);
2578 cxgb4_insert_tid(t, child_ep, hwtid);
2579 insert_ep_tid(child_ep);
2580 if (accept_cr(child_ep, skb, req)) {
2581 c4iw_put_ep(&parent_ep->com);
2582 release_ep_resources(child_ep);
2584 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2587 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2588 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2589 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2593 c4iw_put_ep(&child_ep->com);
2595 reject_cr(dev, hwtid, skb);
2597 c4iw_put_ep(&parent_ep->com);
2602 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2605 struct cpl_pass_establish *req = cplhdr(skb);
2606 unsigned int tid = GET_TID(req);
2609 ep = get_ep_from_tid(dev, tid);
2610 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2611 ep->snd_seq = be32_to_cpu(req->snd_isn);
2612 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2614 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2615 ntohs(req->tcp_opt));
2617 set_emss(ep, ntohs(req->tcp_opt));
2619 dst_confirm(ep->dst);
2620 mutex_lock(&ep->com.mutex);
2621 ep->com.state = MPA_REQ_WAIT;
2623 set_bit(PASS_ESTAB, &ep->com.history);
2624 ret = send_flowc(ep);
2625 mutex_unlock(&ep->com.mutex);
2627 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2628 c4iw_put_ep(&ep->com);
2633 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2635 struct cpl_peer_close *hdr = cplhdr(skb);
2637 struct c4iw_qp_attributes attrs;
2640 unsigned int tid = GET_TID(hdr);
2643 ep = get_ep_from_tid(dev, tid);
2647 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2648 dst_confirm(ep->dst);
2650 set_bit(PEER_CLOSE, &ep->com.history);
2651 mutex_lock(&ep->com.mutex);
2652 switch (ep->com.state) {
2654 __state_set(&ep->com, CLOSING);
2657 __state_set(&ep->com, CLOSING);
2658 connect_reply_upcall(ep, -ECONNRESET);
2663 * We're gonna mark this puppy DEAD, but keep
2664 * the reference on it until the ULP accepts or
2665 * rejects the CR. Also wake up anyone waiting
2666 * in rdma connection migration (see c4iw_accept_cr()).
2668 __state_set(&ep->com, CLOSING);
2669 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2670 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2673 __state_set(&ep->com, CLOSING);
2674 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2675 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2679 __state_set(&ep->com, CLOSING);
2680 attrs.next_state = C4IW_QP_STATE_CLOSING;
2681 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2682 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2683 if (ret != -ECONNRESET) {
2684 peer_close_upcall(ep);
2692 __state_set(&ep->com, MORIBUND);
2696 (void)stop_ep_timer(ep);
2697 if (ep->com.cm_id && ep->com.qp) {
2698 attrs.next_state = C4IW_QP_STATE_IDLE;
2699 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2700 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2702 close_complete_upcall(ep, 0);
2703 __state_set(&ep->com, DEAD);
2713 mutex_unlock(&ep->com.mutex);
2715 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2717 release_ep_resources(ep);
2718 c4iw_put_ep(&ep->com);
2722 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2724 struct cpl_abort_req_rss *req = cplhdr(skb);
2726 struct sk_buff *rpl_skb;
2727 struct c4iw_qp_attributes attrs;
2730 unsigned int tid = GET_TID(req);
2731 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2733 ep = get_ep_from_tid(dev, tid);
2737 if (cxgb_is_neg_adv(req->status)) {
2738 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2739 __func__, ep->hwtid, req->status,
2740 neg_adv_str(req->status));
2741 ep->stats.abort_neg_adv++;
2742 mutex_lock(&dev->rdev.stats.lock);
2743 dev->rdev.stats.neg_adv++;
2744 mutex_unlock(&dev->rdev.stats.lock);
2747 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2749 set_bit(PEER_ABORT, &ep->com.history);
2752 * Wake up any threads in rdma_init() or rdma_fini().
2753 * However, this is not needed if com state is just
2756 if (ep->com.state != MPA_REQ_SENT)
2757 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2759 mutex_lock(&ep->com.mutex);
2760 switch (ep->com.state) {
2762 c4iw_put_ep(&ep->parent_ep->com);
2765 (void)stop_ep_timer(ep);
2768 (void)stop_ep_timer(ep);
2769 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2770 connect_reply_upcall(ep, -ECONNRESET);
2773 * we just don't send notification upwards because we
2774 * want to retry with mpa_v1 without upper layers even
2777 * do some housekeeping so as to re-initiate the
2780 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2782 ep->retry_with_mpa_v1 = 1;
2794 if (ep->com.cm_id && ep->com.qp) {
2795 attrs.next_state = C4IW_QP_STATE_ERROR;
2796 ret = c4iw_modify_qp(ep->com.qp->rhp,
2797 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2801 "%s - qp <- error failed!\n",
2804 peer_abort_upcall(ep);
2809 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2810 mutex_unlock(&ep->com.mutex);
2816 dst_confirm(ep->dst);
2817 if (ep->com.state != ABORTING) {
2818 __state_set(&ep->com, DEAD);
2819 /* we don't release if we want to retry with mpa_v1 */
2820 if (!ep->retry_with_mpa_v1)
2823 mutex_unlock(&ep->com.mutex);
2825 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2826 if (WARN_ON(!rpl_skb)) {
2831 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2833 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2836 release_ep_resources(ep);
2837 else if (ep->retry_with_mpa_v1) {
2838 if (ep->com.remote_addr.ss_family == AF_INET6) {
2839 struct sockaddr_in6 *sin6 =
2840 (struct sockaddr_in6 *)
2841 &ep->com.local_addr;
2843 ep->com.dev->rdev.lldi.ports[0],
2844 (const u32 *)&sin6->sin6_addr.s6_addr,
2847 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2848 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2849 dst_release(ep->dst);
2850 cxgb4_l2t_release(ep->l2t);
2855 c4iw_put_ep(&ep->com);
2856 /* Dereferencing ep, referenced in peer_abort_intr() */
2857 c4iw_put_ep(&ep->com);
2861 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2864 struct c4iw_qp_attributes attrs;
2865 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2867 unsigned int tid = GET_TID(rpl);
2869 ep = get_ep_from_tid(dev, tid);
2873 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2876 /* The cm_id may be null if we failed to connect */
2877 mutex_lock(&ep->com.mutex);
2878 set_bit(CLOSE_CON_RPL, &ep->com.history);
2879 switch (ep->com.state) {
2881 __state_set(&ep->com, MORIBUND);
2884 (void)stop_ep_timer(ep);
2885 if ((ep->com.cm_id) && (ep->com.qp)) {
2886 attrs.next_state = C4IW_QP_STATE_IDLE;
2887 c4iw_modify_qp(ep->com.qp->rhp,
2889 C4IW_QP_ATTR_NEXT_STATE,
2892 close_complete_upcall(ep, 0);
2893 __state_set(&ep->com, DEAD);
2903 mutex_unlock(&ep->com.mutex);
2905 release_ep_resources(ep);
2906 c4iw_put_ep(&ep->com);
2910 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2912 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2913 unsigned int tid = GET_TID(rpl);
2915 struct c4iw_qp_attributes attrs;
2917 ep = get_ep_from_tid(dev, tid);
2920 if (ep && ep->com.qp) {
2921 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2922 ep->com.qp->wq.sq.qid);
2923 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2924 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2925 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2927 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2928 c4iw_put_ep(&ep->com);
2934 * Upcall from the adapter indicating data has been transmitted.
2935 * For us its just the single MPA request or reply. We can now free
2936 * the skb holding the mpa message.
2938 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2941 struct cpl_fw4_ack *hdr = cplhdr(skb);
2942 u8 credits = hdr->credits;
2943 unsigned int tid = GET_TID(hdr);
2946 ep = get_ep_from_tid(dev, tid);
2949 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2951 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2952 __func__, ep, ep->hwtid, state_read(&ep->com));
2956 dst_confirm(ep->dst);
2958 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2959 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2960 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2961 mutex_lock(&ep->com.mutex);
2962 kfree_skb(ep->mpa_skb);
2964 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2966 mutex_unlock(&ep->com.mutex);
2969 c4iw_put_ep(&ep->com);
2973 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2976 struct c4iw_ep *ep = to_ep(cm_id);
2978 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2980 mutex_lock(&ep->com.mutex);
2981 if (ep->com.state != MPA_REQ_RCVD) {
2982 mutex_unlock(&ep->com.mutex);
2983 c4iw_put_ep(&ep->com);
2986 set_bit(ULP_REJECT, &ep->com.history);
2990 abort = send_mpa_reject(ep, pdata, pdata_len);
2991 mutex_unlock(&ep->com.mutex);
2994 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
2995 c4iw_put_ep(&ep->com);
2999 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3002 struct c4iw_qp_attributes attrs;
3003 enum c4iw_qp_attr_mask mask;
3004 struct c4iw_ep *ep = to_ep(cm_id);
3005 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3006 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3009 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3011 mutex_lock(&ep->com.mutex);
3012 if (ep->com.state != MPA_REQ_RCVD) {
3019 set_bit(ULP_ACCEPT, &ep->com.history);
3020 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3021 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3026 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3027 if (conn_param->ord > ep->ird) {
3028 if (RELAXED_IRD_NEGOTIATION) {
3029 conn_param->ord = ep->ird;
3031 ep->ird = conn_param->ird;
3032 ep->ord = conn_param->ord;
3033 send_mpa_reject(ep, conn_param->private_data,
3034 conn_param->private_data_len);
3039 if (conn_param->ird < ep->ord) {
3040 if (RELAXED_IRD_NEGOTIATION &&
3041 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3042 conn_param->ird = ep->ord;
3049 ep->ird = conn_param->ird;
3050 ep->ord = conn_param->ord;
3052 if (ep->mpa_attr.version == 1) {
3053 if (peer2peer && ep->ird == 0)
3057 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3058 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3062 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3064 ep->com.cm_id = cm_id;
3065 ref_cm_id(&ep->com);
3069 /* bind QP to EP and move to RTS */
3070 attrs.mpa_attr = ep->mpa_attr;
3071 attrs.max_ird = ep->ird;
3072 attrs.max_ord = ep->ord;
3073 attrs.llp_stream_handle = ep;
3074 attrs.next_state = C4IW_QP_STATE_RTS;
3076 /* bind QP and TID with INIT_WR */
3077 mask = C4IW_QP_ATTR_NEXT_STATE |
3078 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3079 C4IW_QP_ATTR_MPA_ATTR |
3080 C4IW_QP_ATTR_MAX_IRD |
3081 C4IW_QP_ATTR_MAX_ORD;
3083 err = c4iw_modify_qp(ep->com.qp->rhp,
3084 ep->com.qp, mask, &attrs, 1);
3086 goto err_deref_cm_id;
3088 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3089 err = send_mpa_reply(ep, conn_param->private_data,
3090 conn_param->private_data_len);
3092 goto err_deref_cm_id;
3094 __state_set(&ep->com, FPDU_MODE);
3095 established_upcall(ep);
3096 mutex_unlock(&ep->com.mutex);
3097 c4iw_put_ep(&ep->com);
3100 deref_cm_id(&ep->com);
3104 mutex_unlock(&ep->com.mutex);
3106 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3107 c4iw_put_ep(&ep->com);
3111 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3113 struct in_device *ind;
3115 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3116 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3118 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3120 return -EADDRNOTAVAIL;
3121 for_primary_ifa(ind) {
3122 laddr->sin_addr.s_addr = ifa->ifa_address;
3123 raddr->sin_addr.s_addr = ifa->ifa_address;
3129 return found ? 0 : -EADDRNOTAVAIL;
3132 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3133 unsigned char banned_flags)
3135 struct inet6_dev *idev;
3136 int err = -EADDRNOTAVAIL;
3139 idev = __in6_dev_get(dev);
3141 struct inet6_ifaddr *ifp;
3143 read_lock_bh(&idev->lock);
3144 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3145 if (ifp->scope == IFA_LINK &&
3146 !(ifp->flags & banned_flags)) {
3147 memcpy(addr, &ifp->addr, 16);
3152 read_unlock_bh(&idev->lock);
3158 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3160 struct in6_addr uninitialized_var(addr);
3161 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3162 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3164 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3165 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3166 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3169 return -EADDRNOTAVAIL;
3172 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3174 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3177 struct sockaddr_in *laddr;
3178 struct sockaddr_in *raddr;
3179 struct sockaddr_in6 *laddr6;
3180 struct sockaddr_in6 *raddr6;
3184 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3185 (conn_param->ird > cur_max_read_depth(dev))) {
3189 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3191 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3196 skb_queue_head_init(&ep->com.ep_skb_list);
3197 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3202 init_timer(&ep->timer);
3203 ep->plen = conn_param->private_data_len;
3205 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3206 conn_param->private_data, ep->plen);
3207 ep->ird = conn_param->ird;
3208 ep->ord = conn_param->ord;
3210 if (peer2peer && ep->ord == 0)
3213 ep->com.cm_id = cm_id;
3214 ref_cm_id(&ep->com);
3216 ep->com.qp = get_qhp(dev, conn_param->qpn);
3218 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3223 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3227 * Allocate an active TID to initiate a TCP connection.
3229 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3230 if (ep->atid == -1) {
3231 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3235 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3237 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3238 sizeof(ep->com.local_addr));
3239 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3240 sizeof(ep->com.remote_addr));
3242 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3243 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3244 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3245 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3247 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3249 ra = (__u8 *)&raddr->sin_addr;
3252 * Handle loopback requests to INADDR_ANY.
3254 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3255 err = pick_local_ipaddrs(dev, cm_id);
3261 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3262 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3263 ra, ntohs(raddr->sin_port));
3264 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3265 laddr->sin_addr.s_addr,
3266 raddr->sin_addr.s_addr,
3268 raddr->sin_port, cm_id->tos);
3271 ra = (__u8 *)&raddr6->sin6_addr;
3274 * Handle loopback requests to INADDR_ANY.
3276 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3277 err = pick_local_ip6addrs(dev, cm_id);
3283 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3284 __func__, laddr6->sin6_addr.s6_addr,
3285 ntohs(laddr6->sin6_port),
3286 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3287 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3288 laddr6->sin6_addr.s6_addr,
3289 raddr6->sin6_addr.s6_addr,
3291 raddr6->sin6_port, 0,
3292 raddr6->sin6_scope_id);
3295 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3296 err = -EHOSTUNREACH;
3300 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3301 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3303 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3307 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3308 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3311 state_set(&ep->com, CONNECTING);
3312 ep->tos = cm_id->tos;
3314 /* send connect request to rnic */
3315 err = send_connect(ep);
3319 cxgb4_l2t_release(ep->l2t);
3321 dst_release(ep->dst);
3323 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3324 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3326 skb_queue_purge(&ep->com.ep_skb_list);
3327 deref_cm_id(&ep->com);
3329 c4iw_put_ep(&ep->com);
3334 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3337 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3338 &ep->com.local_addr;
3340 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3341 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3342 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3346 c4iw_init_wr_wait(&ep->com.wr_wait);
3347 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3348 ep->stid, &sin6->sin6_addr,
3350 ep->com.dev->rdev.lldi.rxq_ids[0]);
3352 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3356 err = net_xmit_errno(err);
3358 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3359 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3360 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3362 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3367 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3370 struct sockaddr_in *sin = (struct sockaddr_in *)
3371 &ep->com.local_addr;
3373 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3375 err = cxgb4_create_server_filter(
3376 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3377 sin->sin_addr.s_addr, sin->sin_port, 0,
3378 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3379 if (err == -EBUSY) {
3380 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3384 set_current_state(TASK_UNINTERRUPTIBLE);
3385 schedule_timeout(usecs_to_jiffies(100));
3387 } while (err == -EBUSY);
3389 c4iw_init_wr_wait(&ep->com.wr_wait);
3390 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3391 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3392 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3394 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3398 err = net_xmit_errno(err);
3401 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3403 &sin->sin_addr, ntohs(sin->sin_port));
3407 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3410 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3411 struct c4iw_listen_ep *ep;
3415 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3417 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3421 skb_queue_head_init(&ep->com.ep_skb_list);
3422 PDBG("%s ep %p\n", __func__, ep);
3423 ep->com.cm_id = cm_id;
3424 ref_cm_id(&ep->com);
3426 ep->backlog = backlog;
3427 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3428 sizeof(ep->com.local_addr));
3431 * Allocate a server TID.
3433 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3434 ep->com.local_addr.ss_family == AF_INET)
3435 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3436 cm_id->m_local_addr.ss_family, ep);
3438 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3439 cm_id->m_local_addr.ss_family, ep);
3441 if (ep->stid == -1) {
3442 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3446 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3448 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3449 sizeof(ep->com.local_addr));
3451 state_set(&ep->com, LISTEN);
3452 if (ep->com.local_addr.ss_family == AF_INET)
3453 err = create_server4(dev, ep);
3455 err = create_server6(dev, ep);
3457 cm_id->provider_data = ep;
3461 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3462 ep->com.local_addr.ss_family);
3464 deref_cm_id(&ep->com);
3465 c4iw_put_ep(&ep->com);
3471 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3474 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3476 PDBG("%s ep %p\n", __func__, ep);
3479 state_set(&ep->com, DEAD);
3480 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3481 ep->com.local_addr.ss_family == AF_INET) {
3482 err = cxgb4_remove_server_filter(
3483 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3484 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3486 struct sockaddr_in6 *sin6;
3487 c4iw_init_wr_wait(&ep->com.wr_wait);
3488 err = cxgb4_remove_server(
3489 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3490 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3493 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3495 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3496 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3497 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3499 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3500 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3501 ep->com.local_addr.ss_family);
3503 deref_cm_id(&ep->com);
3504 c4iw_put_ep(&ep->com);
3508 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3513 struct c4iw_rdev *rdev;
3515 mutex_lock(&ep->com.mutex);
3517 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3518 states[ep->com.state], abrupt);
3521 * Ref the ep here in case we have fatal errors causing the
3522 * ep to be released and freed.
3524 c4iw_get_ep(&ep->com);
3526 rdev = &ep->com.dev->rdev;
3527 if (c4iw_fatal_error(rdev)) {
3529 close_complete_upcall(ep, -EIO);
3530 ep->com.state = DEAD;
3532 switch (ep->com.state) {
3541 ep->com.state = ABORTING;
3543 ep->com.state = CLOSING;
3546 * if we close before we see the fw4_ack() then we fix
3547 * up the timer state since we're reusing it.
3550 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3551 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3556 set_bit(CLOSE_SENT, &ep->com.flags);
3559 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3562 (void)stop_ep_timer(ep);
3563 ep->com.state = ABORTING;
3565 ep->com.state = MORIBUND;
3571 PDBG("%s ignoring disconnect ep %p state %u\n",
3572 __func__, ep, ep->com.state);
3581 set_bit(EP_DISC_ABORT, &ep->com.history);
3582 close_complete_upcall(ep, -ECONNRESET);
3583 ret = send_abort(ep);
3585 set_bit(EP_DISC_CLOSE, &ep->com.history);
3586 ret = send_halfclose(ep);
3589 set_bit(EP_DISC_FAIL, &ep->com.history);
3592 close_complete_upcall(ep, -EIO);
3595 struct c4iw_qp_attributes attrs;
3597 attrs.next_state = C4IW_QP_STATE_ERROR;
3598 ret = c4iw_modify_qp(ep->com.qp->rhp,
3600 C4IW_QP_ATTR_NEXT_STATE,
3604 "%s - qp <- error failed!\n",
3610 mutex_unlock(&ep->com.mutex);
3611 c4iw_put_ep(&ep->com);
3613 release_ep_resources(ep);
3617 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3618 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3621 int atid = be32_to_cpu(req->tid);
3623 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3624 (__force u32) req->tid);
3628 switch (req->retval) {
3630 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3631 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3632 send_fw_act_open_req(ep, atid);
3636 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3637 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3638 send_fw_act_open_req(ep, atid);
3643 pr_info("%s unexpected ofld conn wr retval %d\n",
3644 __func__, req->retval);
3647 pr_err("active ofld_connect_wr failure %d atid %d\n",
3649 mutex_lock(&dev->rdev.stats.lock);
3650 dev->rdev.stats.act_ofld_conn_fails++;
3651 mutex_unlock(&dev->rdev.stats.lock);
3652 connect_reply_upcall(ep, status2errno(req->retval));
3653 state_set(&ep->com, DEAD);
3654 if (ep->com.remote_addr.ss_family == AF_INET6) {
3655 struct sockaddr_in6 *sin6 =
3656 (struct sockaddr_in6 *)&ep->com.local_addr;
3657 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3658 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3660 remove_handle(dev, &dev->atid_idr, atid);
3661 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3662 dst_release(ep->dst);
3663 cxgb4_l2t_release(ep->l2t);
3664 c4iw_put_ep(&ep->com);
3667 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3668 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3670 struct sk_buff *rpl_skb;
3671 struct cpl_pass_accept_req *cpl;
3674 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3677 PDBG("%s passive open failure %d\n", __func__, req->retval);
3678 mutex_lock(&dev->rdev.stats.lock);
3679 dev->rdev.stats.pas_ofld_conn_fails++;
3680 mutex_unlock(&dev->rdev.stats.lock);
3683 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3684 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3685 (__force u32) htonl(
3686 (__force u32) req->tid)));
3687 ret = pass_accept_req(dev, rpl_skb);
3694 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3696 struct cpl_fw6_msg *rpl = cplhdr(skb);
3697 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3699 switch (rpl->type) {
3701 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3703 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3704 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3705 switch (req->t_state) {
3707 active_ofld_conn_reply(dev, skb, req);
3710 passive_ofld_conn_reply(dev, skb, req);
3713 pr_err("%s unexpected ofld conn wr state %d\n",
3714 __func__, req->t_state);
3722 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3725 __be16 hdr_len, vlantag, len;
3727 int tcp_hdr_len, ip_hdr_len;
3729 struct cpl_rx_pkt *cpl = cplhdr(skb);
3730 struct cpl_pass_accept_req *req;
3731 struct tcp_options_received tmp_opt;
3732 struct c4iw_dev *dev;
3733 enum chip_type type;
3735 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3736 /* Store values from cpl_rx_pkt in temporary location. */
3737 vlantag = cpl->vlan;
3739 l2info = cpl->l2info;
3740 hdr_len = cpl->hdr_len;
3743 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3746 * We need to parse the TCP options from SYN packet.
3747 * to generate cpl_pass_accept_req.
3749 memset(&tmp_opt, 0, sizeof(tmp_opt));
3750 tcp_clear_options(&tmp_opt);
3751 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3753 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3754 memset(req, 0, sizeof(*req));
3755 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3756 SYN_MAC_IDX_V(RX_MACIDX_G(
3757 be32_to_cpu(l2info))) |
3759 type = dev->rdev.lldi.adapter_type;
3760 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3761 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3763 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3764 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3765 eth_hdr_len = is_t4(type) ?
3766 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3767 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3768 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3769 IP_HDR_LEN_V(ip_hdr_len) |
3770 ETH_HDR_LEN_V(eth_hdr_len));
3771 } else { /* T6 and later */
3772 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3773 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3774 T6_IP_HDR_LEN_V(ip_hdr_len) |
3775 T6_ETH_HDR_LEN_V(eth_hdr_len));
3777 req->vlan = vlantag;
3779 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3780 PASS_OPEN_TOS_V(tos));
3781 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3782 if (tmp_opt.wscale_ok)
3783 req->tcpopt.wsf = tmp_opt.snd_wscale;
3784 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3785 if (tmp_opt.sack_ok)
3786 req->tcpopt.sack = 1;
3787 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3791 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3792 __be32 laddr, __be16 lport,
3793 __be32 raddr, __be16 rport,
3794 u32 rcv_isn, u32 filter, u16 window,
3795 u32 rss_qid, u8 port_id)
3797 struct sk_buff *req_skb;
3798 struct fw_ofld_connection_wr *req;
3799 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3802 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3803 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3804 memset(req, 0, sizeof(*req));
3805 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3806 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3807 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3808 req->le.filter = (__force __be32) filter;
3809 req->le.lport = lport;
3810 req->le.pport = rport;
3811 req->le.u.ipv4.lip = laddr;
3812 req->le.u.ipv4.pip = raddr;
3813 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3814 req->tcb.rcv_adv = htons(window);
3815 req->tcb.t_state_to_astid =
3816 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3817 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3818 FW_OFLD_CONNECTION_WR_ASTID_V(
3819 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3822 * We store the qid in opt2 which will be used by the firmware
3823 * to send us the wr response.
3825 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3828 * We initialize the MSS index in TCB to 0xF.
3829 * So that when driver sends cpl_pass_accept_rpl
3830 * TCB picks up the correct value. If this was 0
3831 * TP will ignore any value > 0 for MSS index.
3833 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3834 req->cookie = (uintptr_t)skb;
3836 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3837 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3839 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3847 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3848 * messages when a filter is being used instead of server to
3849 * redirect a syn packet. When packets hit filter they are redirected
3850 * to the offload queue and driver tries to establish the connection
3851 * using firmware work request.
3853 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3856 unsigned int filter;
3857 struct ethhdr *eh = NULL;
3858 struct vlan_ethhdr *vlan_eh = NULL;
3860 struct tcphdr *tcph;
3861 struct rss_header *rss = (void *)skb->data;
3862 struct cpl_rx_pkt *cpl = (void *)skb->data;
3863 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3864 struct l2t_entry *e;
3865 struct dst_entry *dst;
3866 struct c4iw_ep *lep = NULL;
3868 struct port_info *pi;
3869 struct net_device *pdev;
3870 u16 rss_qid, eth_hdr_len;
3873 struct neighbour *neigh;
3875 /* Drop all non-SYN packets */
3876 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3880 * Drop all packets which did not hit the filter.
3881 * Unlikely to happen.
3883 if (!(rss->filter_hit && rss->filter_tid))
3887 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3889 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3891 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3893 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3897 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3899 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3902 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3905 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3908 pr_err("T%d Chip is not supported\n",
3909 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3913 if (eth_hdr_len == ETH_HLEN) {
3914 eh = (struct ethhdr *)(req + 1);
3915 iph = (struct iphdr *)(eh + 1);
3917 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3918 iph = (struct iphdr *)(vlan_eh + 1);
3919 skb->vlan_tci = ntohs(cpl->vlan);
3922 if (iph->version != 0x4)
3925 tcph = (struct tcphdr *)(iph + 1);
3926 skb_set_network_header(skb, (void *)iph - (void *)rss);
3927 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3930 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3931 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3932 ntohs(tcph->source), iph->tos);
3934 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3935 iph->daddr, iph->saddr, tcph->dest,
3936 tcph->source, iph->tos);
3938 pr_err("%s - failed to find dst entry!\n",
3942 neigh = dst_neigh_lookup_skb(dst, skb);
3945 pr_err("%s - failed to allocate neigh!\n",
3950 if (neigh->dev->flags & IFF_LOOPBACK) {
3951 pdev = ip_dev_find(&init_net, iph->daddr);
3952 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3954 pi = (struct port_info *)netdev_priv(pdev);
3955 tx_chan = cxgb4_port_chan(pdev);
3958 pdev = get_real_dev(neigh->dev);
3959 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3961 pi = (struct port_info *)netdev_priv(pdev);
3962 tx_chan = cxgb4_port_chan(pdev);
3964 neigh_release(neigh);
3966 pr_err("%s - failed to allocate l2t entry!\n",
3971 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3972 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3973 window = (__force u16) htons((__force u16)tcph->window);
3975 /* Calcuate filter portion for LE region. */
3976 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3977 dev->rdev.lldi.ports[0],
3981 * Synthesize the cpl_pass_accept_req. We have everything except the
3982 * TID. Once firmware sends a reply with TID we update the TID field
3983 * in cpl and pass it through the regular cpl_pass_accept_req path.
3985 build_cpl_pass_accept_req(skb, stid, iph->tos);
3986 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3987 tcph->source, ntohl(tcph->seq), filter, window,
3988 rss_qid, pi->port_id);
3989 cxgb4_l2t_release(e);
3994 c4iw_put_ep(&lep->com);
3999 * These are the real handlers that are called from a
4002 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4003 [CPL_ACT_ESTABLISH] = act_establish,
4004 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4005 [CPL_RX_DATA] = rx_data,
4006 [CPL_ABORT_RPL_RSS] = abort_rpl,
4007 [CPL_ABORT_RPL] = abort_rpl,
4008 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4009 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4010 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4011 [CPL_PASS_ESTABLISH] = pass_establish,
4012 [CPL_PEER_CLOSE] = peer_close,
4013 [CPL_ABORT_REQ_RSS] = peer_abort,
4014 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4015 [CPL_RDMA_TERMINATE] = terminate,
4016 [CPL_FW4_ACK] = fw4_ack,
4017 [CPL_FW6_MSG] = deferred_fw6_msg,
4018 [CPL_RX_PKT] = rx_pkt,
4019 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4020 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4023 static void process_timeout(struct c4iw_ep *ep)
4025 struct c4iw_qp_attributes attrs;
4028 mutex_lock(&ep->com.mutex);
4029 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4031 set_bit(TIMEDOUT, &ep->com.history);
4032 switch (ep->com.state) {
4034 connect_reply_upcall(ep, -ETIMEDOUT);
4043 if (ep->com.cm_id && ep->com.qp) {
4044 attrs.next_state = C4IW_QP_STATE_ERROR;
4045 c4iw_modify_qp(ep->com.qp->rhp,
4046 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4049 close_complete_upcall(ep, -ETIMEDOUT);
4055 * These states are expected if the ep timed out at the same
4056 * time as another thread was calling stop_ep_timer().
4057 * So we silently do nothing for these states.
4062 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4063 __func__, ep, ep->hwtid, ep->com.state);
4066 mutex_unlock(&ep->com.mutex);
4068 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4069 c4iw_put_ep(&ep->com);
4072 static void process_timedout_eps(void)
4076 spin_lock_irq(&timeout_lock);
4077 while (!list_empty(&timeout_list)) {
4078 struct list_head *tmp;
4080 tmp = timeout_list.next;
4084 spin_unlock_irq(&timeout_lock);
4085 ep = list_entry(tmp, struct c4iw_ep, entry);
4086 process_timeout(ep);
4087 spin_lock_irq(&timeout_lock);
4089 spin_unlock_irq(&timeout_lock);
4092 static void process_work(struct work_struct *work)
4094 struct sk_buff *skb = NULL;
4095 struct c4iw_dev *dev;
4096 struct cpl_act_establish *rpl;
4097 unsigned int opcode;
4100 process_timedout_eps();
4101 while ((skb = skb_dequeue(&rxq))) {
4103 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4104 opcode = rpl->ot.opcode;
4106 BUG_ON(!work_handlers[opcode]);
4107 ret = work_handlers[opcode](dev, skb);
4110 process_timedout_eps();
4114 static DECLARE_WORK(skb_work, process_work);
4116 static void ep_timeout(unsigned long arg)
4118 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4121 spin_lock(&timeout_lock);
4122 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4124 * Only insert if it is not already on the list.
4126 if (!ep->entry.next) {
4127 list_add_tail(&ep->entry, &timeout_list);
4131 spin_unlock(&timeout_lock);
4133 queue_work(workq, &skb_work);
4137 * All the CM events are handled on a work queue to have a safe context.
4139 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4143 * Save dev in the skb->cb area.
4145 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4148 * Queue the skb and schedule the worker thread.
4150 skb_queue_tail(&rxq, skb);
4151 queue_work(workq, &skb_work);
4155 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4157 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4159 if (rpl->status != CPL_ERR_NONE) {
4160 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4161 "for tid %u\n", rpl->status, GET_TID(rpl));
4167 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4169 struct cpl_fw6_msg *rpl = cplhdr(skb);
4170 struct c4iw_wr_wait *wr_waitp;
4173 PDBG("%s type %u\n", __func__, rpl->type);
4175 switch (rpl->type) {
4176 case FW6_TYPE_WR_RPL:
4177 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4178 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4179 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4181 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4185 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4189 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4197 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4199 struct cpl_abort_req_rss *req = cplhdr(skb);
4201 unsigned int tid = GET_TID(req);
4203 ep = get_ep_from_tid(dev, tid);
4204 /* This EP will be dereferenced in peer_abort() */
4206 printk(KERN_WARNING MOD
4207 "Abort on non-existent endpoint, tid %d\n", tid);
4211 if (cxgb_is_neg_adv(req->status)) {
4212 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4213 __func__, ep->hwtid, req->status,
4214 neg_adv_str(req->status));
4217 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4220 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4227 * Most upcalls from the T4 Core go to sched() to
4228 * schedule the processing on a work queue.
4230 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4231 [CPL_ACT_ESTABLISH] = sched,
4232 [CPL_ACT_OPEN_RPL] = sched,
4233 [CPL_RX_DATA] = sched,
4234 [CPL_ABORT_RPL_RSS] = sched,
4235 [CPL_ABORT_RPL] = sched,
4236 [CPL_PASS_OPEN_RPL] = sched,
4237 [CPL_CLOSE_LISTSRV_RPL] = sched,
4238 [CPL_PASS_ACCEPT_REQ] = sched,
4239 [CPL_PASS_ESTABLISH] = sched,
4240 [CPL_PEER_CLOSE] = sched,
4241 [CPL_CLOSE_CON_RPL] = sched,
4242 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4243 [CPL_RDMA_TERMINATE] = sched,
4244 [CPL_FW4_ACK] = sched,
4245 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4246 [CPL_FW6_MSG] = fw6_msg,
4247 [CPL_RX_PKT] = sched
4250 int __init c4iw_cm_init(void)
4252 spin_lock_init(&timeout_lock);
4253 skb_queue_head_init(&rxq);
4255 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4262 void c4iw_cm_term(void)
4264 WARN_ON(!list_empty(&timeout_list));
4265 flush_workqueue(workq);
4266 destroy_workqueue(workq);