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, "obsolete");
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 pr_debug("%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 pr_debug("%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 pr_debug("%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 pr_debug("%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 pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n",
267 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
268 pr_debug("%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 pr_debug("%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 pr_debug("%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 pr_debug("%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("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);
495 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
499 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
500 c4iw_put_ep(&ep->parent_ep->com);
501 release_ep_resources(ep);
507 * Fake up a special CPL opcode and call sched() so process_work() will call
508 * _put_ep_safe() in a safe context to free the ep resources. This is needed
509 * because ARP error handlers are called in an ATOMIC context, and
510 * _c4iw_free_ep() needs to block.
512 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
515 struct cpl_act_establish *rpl = cplhdr(skb);
517 /* Set our special ARP_FAILURE opcode */
518 rpl->ot.opcode = cpl;
521 * Save ep in the skb->cb area, after where sched() will save the dev
524 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
525 sched(ep->com.dev, skb);
528 /* Handle an ARP failure for an accept */
529 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
531 struct c4iw_ep *ep = handle;
533 pr_err("ARP failure during accept - tid %u - dropping connection\n",
536 __state_set(&ep->com, DEAD);
537 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
541 * Handle an ARP failure for an active open.
543 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
545 struct c4iw_ep *ep = handle;
547 pr_err("ARP failure during connect\n");
548 connect_reply_upcall(ep, -EHOSTUNREACH);
549 __state_set(&ep->com, DEAD);
550 if (ep->com.remote_addr.ss_family == AF_INET6) {
551 struct sockaddr_in6 *sin6 =
552 (struct sockaddr_in6 *)&ep->com.local_addr;
553 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
554 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
556 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
557 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
558 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
562 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
565 static void abort_arp_failure(void *handle, struct sk_buff *skb)
568 struct c4iw_ep *ep = handle;
569 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
570 struct cpl_abort_req *req = cplhdr(skb);
572 pr_debug("%s rdev %p\n", __func__, rdev);
573 req->cmd = CPL_ABORT_NO_RST;
575 ret = c4iw_ofld_send(rdev, skb);
577 __state_set(&ep->com, DEAD);
578 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
583 static int send_flowc(struct c4iw_ep *ep)
585 struct fw_flowc_wr *flowc;
586 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
588 u16 vlan = ep->l2t->vlan;
594 if (vlan == CPL_L2T_VLAN_NONE)
599 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
601 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
602 FW_FLOWC_WR_NPARAMS_V(nparams));
603 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
604 16)) | FW_WR_FLOWID_V(ep->hwtid));
606 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
607 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
608 (ep->com.dev->rdev.lldi.pf));
609 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
610 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
611 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
612 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
613 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
614 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
615 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
616 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
617 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
618 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
619 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
620 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
621 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
622 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
626 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
627 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
628 flowc->mnemval[8].val = cpu_to_be32(pri);
630 /* Pad WR to 16 byte boundary */
631 flowc->mnemval[8].mnemonic = 0;
632 flowc->mnemval[8].val = 0;
634 for (i = 0; i < 9; i++) {
635 flowc->mnemval[i].r4[0] = 0;
636 flowc->mnemval[i].r4[1] = 0;
637 flowc->mnemval[i].r4[2] = 0;
640 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
641 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
644 static int send_halfclose(struct c4iw_ep *ep)
646 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
647 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
649 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
653 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
654 NULL, arp_failure_discard);
656 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
659 static int send_abort(struct c4iw_ep *ep)
661 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
662 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
664 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
665 if (WARN_ON(!req_skb))
668 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
669 ep, abort_arp_failure);
671 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
674 static int send_connect(struct c4iw_ep *ep)
676 struct cpl_act_open_req *req = NULL;
677 struct cpl_t5_act_open_req *t5req = NULL;
678 struct cpl_t6_act_open_req *t6req = NULL;
679 struct cpl_act_open_req6 *req6 = NULL;
680 struct cpl_t5_act_open_req6 *t5req6 = NULL;
681 struct cpl_t6_act_open_req6 *t6req6 = NULL;
685 unsigned int mtu_idx;
687 int win, sizev4, sizev6, wrlen;
688 struct sockaddr_in *la = (struct sockaddr_in *)
690 struct sockaddr_in *ra = (struct sockaddr_in *)
691 &ep->com.remote_addr;
692 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
694 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
695 &ep->com.remote_addr;
697 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
698 u32 isn = (prandom_u32() & ~7UL) - 1;
699 struct net_device *netdev;
702 netdev = ep->com.dev->rdev.lldi.ports[0];
704 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
706 sizev4 = sizeof(struct cpl_act_open_req);
707 sizev6 = sizeof(struct cpl_act_open_req6);
710 sizev4 = sizeof(struct cpl_t5_act_open_req);
711 sizev6 = sizeof(struct cpl_t5_act_open_req6);
714 sizev4 = sizeof(struct cpl_t6_act_open_req);
715 sizev6 = sizeof(struct cpl_t6_act_open_req6);
718 pr_err("T%d Chip is not supported\n",
719 CHELSIO_CHIP_VERSION(adapter_type));
723 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
724 roundup(sizev4, 16) :
727 pr_debug("%s ep %p atid %u\n", __func__, ep, ep->atid);
729 skb = get_skb(NULL, wrlen, GFP_KERNEL);
731 pr_err("%s - failed to alloc skb\n", __func__);
734 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
736 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
737 enable_tcp_timestamps,
738 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
739 wscale = cxgb_compute_wscale(rcv_win);
742 * Specify the largest window that will fit in opt0. The
743 * remainder will be specified in the rx_data_ack.
745 win = ep->rcv_win >> 10;
746 if (win > RCV_BUFSIZ_M)
749 opt0 = (nocong ? NO_CONG_F : 0) |
752 WND_SCALE_V(wscale) |
754 L2T_IDX_V(ep->l2t->idx) |
755 TX_CHAN_V(ep->tx_chan) |
756 SMAC_SEL_V(ep->smac_idx) |
757 DSCP_V(ep->tos >> 2) |
758 ULP_MODE_V(ULP_MODE_TCPDDP) |
760 opt2 = RX_CHANNEL_V(0) |
761 CCTRL_ECN_V(enable_ecn) |
762 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
763 if (enable_tcp_timestamps)
764 opt2 |= TSTAMPS_EN_F;
767 if (wscale && enable_tcp_window_scaling)
768 opt2 |= WND_SCALE_EN_F;
769 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
773 opt2 |= T5_OPT_2_VALID_F;
774 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
778 params = cxgb4_select_ntuple(netdev, ep->l2t);
780 if (ep->com.remote_addr.ss_family == AF_INET6)
781 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
782 (const u32 *)&la6->sin6_addr.s6_addr, 1);
784 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
786 if (ep->com.remote_addr.ss_family == AF_INET) {
787 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
789 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
793 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
795 INIT_TP_WR(t5req, 0);
796 req = (struct cpl_act_open_req *)t5req;
799 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
801 INIT_TP_WR(t6req, 0);
802 req = (struct cpl_act_open_req *)t6req;
803 t5req = (struct cpl_t5_act_open_req *)t6req;
806 pr_err("T%d Chip is not supported\n",
807 CHELSIO_CHIP_VERSION(adapter_type));
812 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
813 ((ep->rss_qid<<14) | ep->atid)));
814 req->local_port = la->sin_port;
815 req->peer_port = ra->sin_port;
816 req->local_ip = la->sin_addr.s_addr;
817 req->peer_ip = ra->sin_addr.s_addr;
818 req->opt0 = cpu_to_be64(opt0);
820 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
821 req->params = cpu_to_be32(params);
822 req->opt2 = cpu_to_be32(opt2);
824 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
826 cpu_to_be64(FILTER_TUPLE_V(params));
827 t5req->rsvd = cpu_to_be32(isn);
828 pr_debug("%s snd_isn %u\n", __func__, t5req->rsvd);
829 t5req->opt2 = cpu_to_be32(opt2);
832 cpu_to_be64(FILTER_TUPLE_V(params));
833 t6req->rsvd = cpu_to_be32(isn);
834 pr_debug("%s snd_isn %u\n", __func__, t6req->rsvd);
835 t6req->opt2 = cpu_to_be32(opt2);
839 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
841 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
845 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
847 INIT_TP_WR(t5req6, 0);
848 req6 = (struct cpl_act_open_req6 *)t5req6;
851 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
853 INIT_TP_WR(t6req6, 0);
854 req6 = (struct cpl_act_open_req6 *)t6req6;
855 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
858 pr_err("T%d Chip is not supported\n",
859 CHELSIO_CHIP_VERSION(adapter_type));
864 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
865 ((ep->rss_qid<<14)|ep->atid)));
866 req6->local_port = la6->sin6_port;
867 req6->peer_port = ra6->sin6_port;
868 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
869 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
870 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
871 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
872 req6->opt0 = cpu_to_be64(opt0);
874 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
875 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
877 req6->opt2 = cpu_to_be32(opt2);
879 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
881 cpu_to_be64(FILTER_TUPLE_V(params));
882 t5req6->rsvd = cpu_to_be32(isn);
883 pr_debug("%s snd_isn %u\n", __func__, t5req6->rsvd);
884 t5req6->opt2 = cpu_to_be32(opt2);
887 cpu_to_be64(FILTER_TUPLE_V(params));
888 t6req6->rsvd = cpu_to_be32(isn);
889 pr_debug("%s snd_isn %u\n", __func__, t6req6->rsvd);
890 t6req6->opt2 = cpu_to_be32(opt2);
896 set_bit(ACT_OPEN_REQ, &ep->com.history);
897 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
899 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
900 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
901 (const u32 *)&la6->sin6_addr.s6_addr, 1);
905 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
908 int mpalen, wrlen, ret;
909 struct fw_ofld_tx_data_wr *req;
910 struct mpa_message *mpa;
911 struct mpa_v2_conn_params mpa_v2_params;
913 pr_debug("%s ep %p tid %u pd_len %d\n",
914 __func__, ep, ep->hwtid, ep->plen);
916 BUG_ON(skb_cloned(skb));
918 mpalen = sizeof(*mpa) + ep->plen;
919 if (mpa_rev_to_use == 2)
920 mpalen += sizeof(struct mpa_v2_conn_params);
921 wrlen = roundup(mpalen + sizeof *req, 16);
922 skb = get_skb(skb, wrlen, GFP_KERNEL);
924 connect_reply_upcall(ep, -ENOMEM);
927 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
929 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
930 memset(req, 0, wrlen);
931 req->op_to_immdlen = cpu_to_be32(
932 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
934 FW_WR_IMMDLEN_V(mpalen));
935 req->flowid_len16 = cpu_to_be32(
936 FW_WR_FLOWID_V(ep->hwtid) |
937 FW_WR_LEN16_V(wrlen >> 4));
938 req->plen = cpu_to_be32(mpalen);
939 req->tunnel_to_proxy = cpu_to_be32(
940 FW_OFLD_TX_DATA_WR_FLUSH_F |
941 FW_OFLD_TX_DATA_WR_SHOVE_F);
943 mpa = (struct mpa_message *)(req + 1);
944 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
948 mpa->flags |= MPA_CRC;
949 if (markers_enabled) {
950 mpa->flags |= MPA_MARKERS;
951 ep->mpa_attr.recv_marker_enabled = 1;
953 ep->mpa_attr.recv_marker_enabled = 0;
955 if (mpa_rev_to_use == 2)
956 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
958 mpa->private_data_size = htons(ep->plen);
959 mpa->revision = mpa_rev_to_use;
960 if (mpa_rev_to_use == 1) {
961 ep->tried_with_mpa_v1 = 1;
962 ep->retry_with_mpa_v1 = 0;
965 if (mpa_rev_to_use == 2) {
966 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
967 sizeof (struct mpa_v2_conn_params));
968 pr_debug("%s initiator ird %u ord %u\n", __func__, ep->ird,
970 mpa_v2_params.ird = htons((u16)ep->ird);
971 mpa_v2_params.ord = htons((u16)ep->ord);
974 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
975 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
977 htons(MPA_V2_RDMA_WRITE_RTR);
978 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
980 htons(MPA_V2_RDMA_READ_RTR);
982 memcpy(mpa->private_data, &mpa_v2_params,
983 sizeof(struct mpa_v2_conn_params));
986 memcpy(mpa->private_data +
987 sizeof(struct mpa_v2_conn_params),
988 ep->mpa_pkt + sizeof(*mpa), ep->plen);
991 memcpy(mpa->private_data,
992 ep->mpa_pkt + sizeof(*mpa), ep->plen);
995 * Reference the mpa skb. This ensures the data area
996 * will remain in memory until the hw acks the tx.
997 * Function fw4_ack() will deref it.
1000 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1001 BUG_ON(ep->mpa_skb);
1003 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1007 __state_set(&ep->com, MPA_REQ_SENT);
1008 ep->mpa_attr.initiator = 1;
1009 ep->snd_seq += mpalen;
1013 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1016 struct fw_ofld_tx_data_wr *req;
1017 struct mpa_message *mpa;
1018 struct sk_buff *skb;
1019 struct mpa_v2_conn_params mpa_v2_params;
1021 pr_debug("%s ep %p tid %u pd_len %d\n",
1022 __func__, ep, ep->hwtid, ep->plen);
1024 mpalen = sizeof(*mpa) + plen;
1025 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1026 mpalen += sizeof(struct mpa_v2_conn_params);
1027 wrlen = roundup(mpalen + sizeof *req, 16);
1029 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1031 pr_err("%s - cannot alloc skb!\n", __func__);
1034 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1036 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1037 memset(req, 0, wrlen);
1038 req->op_to_immdlen = cpu_to_be32(
1039 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1041 FW_WR_IMMDLEN_V(mpalen));
1042 req->flowid_len16 = cpu_to_be32(
1043 FW_WR_FLOWID_V(ep->hwtid) |
1044 FW_WR_LEN16_V(wrlen >> 4));
1045 req->plen = cpu_to_be32(mpalen);
1046 req->tunnel_to_proxy = cpu_to_be32(
1047 FW_OFLD_TX_DATA_WR_FLUSH_F |
1048 FW_OFLD_TX_DATA_WR_SHOVE_F);
1050 mpa = (struct mpa_message *)(req + 1);
1051 memset(mpa, 0, sizeof(*mpa));
1052 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1053 mpa->flags = MPA_REJECT;
1054 mpa->revision = ep->mpa_attr.version;
1055 mpa->private_data_size = htons(plen);
1057 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1058 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1059 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1060 sizeof (struct mpa_v2_conn_params));
1061 mpa_v2_params.ird = htons(((u16)ep->ird) |
1062 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1064 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1066 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1067 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1068 FW_RI_INIT_P2PTYPE_READ_REQ ?
1069 MPA_V2_RDMA_READ_RTR : 0) : 0));
1070 memcpy(mpa->private_data, &mpa_v2_params,
1071 sizeof(struct mpa_v2_conn_params));
1074 memcpy(mpa->private_data +
1075 sizeof(struct mpa_v2_conn_params), pdata, plen);
1078 memcpy(mpa->private_data, pdata, plen);
1081 * Reference the mpa skb again. This ensures the data area
1082 * will remain in memory until the hw acks the tx.
1083 * Function fw4_ack() will deref it.
1086 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1087 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1088 BUG_ON(ep->mpa_skb);
1090 ep->snd_seq += mpalen;
1091 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1094 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1097 struct fw_ofld_tx_data_wr *req;
1098 struct mpa_message *mpa;
1099 struct sk_buff *skb;
1100 struct mpa_v2_conn_params mpa_v2_params;
1102 pr_debug("%s ep %p tid %u pd_len %d\n",
1103 __func__, ep, ep->hwtid, ep->plen);
1105 mpalen = sizeof(*mpa) + plen;
1106 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1107 mpalen += sizeof(struct mpa_v2_conn_params);
1108 wrlen = roundup(mpalen + sizeof *req, 16);
1110 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1112 pr_err("%s - cannot alloc skb!\n", __func__);
1115 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1117 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1118 memset(req, 0, wrlen);
1119 req->op_to_immdlen = cpu_to_be32(
1120 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1122 FW_WR_IMMDLEN_V(mpalen));
1123 req->flowid_len16 = cpu_to_be32(
1124 FW_WR_FLOWID_V(ep->hwtid) |
1125 FW_WR_LEN16_V(wrlen >> 4));
1126 req->plen = cpu_to_be32(mpalen);
1127 req->tunnel_to_proxy = cpu_to_be32(
1128 FW_OFLD_TX_DATA_WR_FLUSH_F |
1129 FW_OFLD_TX_DATA_WR_SHOVE_F);
1131 mpa = (struct mpa_message *)(req + 1);
1132 memset(mpa, 0, sizeof(*mpa));
1133 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1135 if (ep->mpa_attr.crc_enabled)
1136 mpa->flags |= MPA_CRC;
1137 if (ep->mpa_attr.recv_marker_enabled)
1138 mpa->flags |= MPA_MARKERS;
1139 mpa->revision = ep->mpa_attr.version;
1140 mpa->private_data_size = htons(plen);
1142 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1143 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1144 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1145 sizeof (struct mpa_v2_conn_params));
1146 mpa_v2_params.ird = htons((u16)ep->ird);
1147 mpa_v2_params.ord = htons((u16)ep->ord);
1148 if (peer2peer && (ep->mpa_attr.p2p_type !=
1149 FW_RI_INIT_P2PTYPE_DISABLED)) {
1150 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1152 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1153 mpa_v2_params.ord |=
1154 htons(MPA_V2_RDMA_WRITE_RTR);
1155 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1156 mpa_v2_params.ord |=
1157 htons(MPA_V2_RDMA_READ_RTR);
1160 memcpy(mpa->private_data, &mpa_v2_params,
1161 sizeof(struct mpa_v2_conn_params));
1164 memcpy(mpa->private_data +
1165 sizeof(struct mpa_v2_conn_params), pdata, plen);
1168 memcpy(mpa->private_data, pdata, plen);
1171 * Reference the mpa skb. This ensures the data area
1172 * will remain in memory until the hw acks the tx.
1173 * Function fw4_ack() will deref it.
1176 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1178 __state_set(&ep->com, MPA_REP_SENT);
1179 ep->snd_seq += mpalen;
1180 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1183 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1186 struct cpl_act_establish *req = cplhdr(skb);
1187 unsigned int tid = GET_TID(req);
1188 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1189 struct tid_info *t = dev->rdev.lldi.tids;
1192 ep = lookup_atid(t, atid);
1194 pr_debug("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1195 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1197 mutex_lock(&ep->com.mutex);
1198 dst_confirm(ep->dst);
1200 /* setup the hwtid for this connection */
1202 cxgb4_insert_tid(t, ep, tid);
1205 ep->snd_seq = be32_to_cpu(req->snd_isn);
1206 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1208 set_emss(ep, ntohs(req->tcp_opt));
1210 /* dealloc the atid */
1211 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1212 cxgb4_free_atid(t, atid);
1213 set_bit(ACT_ESTAB, &ep->com.history);
1215 /* start MPA negotiation */
1216 ret = send_flowc(ep);
1219 if (ep->retry_with_mpa_v1)
1220 ret = send_mpa_req(ep, skb, 1);
1222 ret = send_mpa_req(ep, skb, mpa_rev);
1225 mutex_unlock(&ep->com.mutex);
1228 mutex_unlock(&ep->com.mutex);
1229 connect_reply_upcall(ep, -ENOMEM);
1230 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1234 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1236 struct iw_cm_event event;
1238 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1239 memset(&event, 0, sizeof(event));
1240 event.event = IW_CM_EVENT_CLOSE;
1241 event.status = status;
1242 if (ep->com.cm_id) {
1243 pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1244 ep, ep->com.cm_id, ep->hwtid);
1245 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1246 deref_cm_id(&ep->com);
1247 set_bit(CLOSE_UPCALL, &ep->com.history);
1251 static void peer_close_upcall(struct c4iw_ep *ep)
1253 struct iw_cm_event event;
1255 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1256 memset(&event, 0, sizeof(event));
1257 event.event = IW_CM_EVENT_DISCONNECT;
1258 if (ep->com.cm_id) {
1259 pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1260 ep, ep->com.cm_id, ep->hwtid);
1261 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1262 set_bit(DISCONN_UPCALL, &ep->com.history);
1266 static void peer_abort_upcall(struct c4iw_ep *ep)
1268 struct iw_cm_event event;
1270 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1271 memset(&event, 0, sizeof(event));
1272 event.event = IW_CM_EVENT_CLOSE;
1273 event.status = -ECONNRESET;
1274 if (ep->com.cm_id) {
1275 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1276 ep->com.cm_id, ep->hwtid);
1277 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1278 deref_cm_id(&ep->com);
1279 set_bit(ABORT_UPCALL, &ep->com.history);
1283 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1285 struct iw_cm_event event;
1287 pr_debug("%s ep %p tid %u status %d\n",
1288 __func__, ep, ep->hwtid, status);
1289 memset(&event, 0, sizeof(event));
1290 event.event = IW_CM_EVENT_CONNECT_REPLY;
1291 event.status = status;
1292 memcpy(&event.local_addr, &ep->com.local_addr,
1293 sizeof(ep->com.local_addr));
1294 memcpy(&event.remote_addr, &ep->com.remote_addr,
1295 sizeof(ep->com.remote_addr));
1297 if ((status == 0) || (status == -ECONNREFUSED)) {
1298 if (!ep->tried_with_mpa_v1) {
1299 /* this means MPA_v2 is used */
1300 event.ord = ep->ird;
1301 event.ird = ep->ord;
1302 event.private_data_len = ep->plen -
1303 sizeof(struct mpa_v2_conn_params);
1304 event.private_data = ep->mpa_pkt +
1305 sizeof(struct mpa_message) +
1306 sizeof(struct mpa_v2_conn_params);
1308 /* this means MPA_v1 is used */
1309 event.ord = cur_max_read_depth(ep->com.dev);
1310 event.ird = cur_max_read_depth(ep->com.dev);
1311 event.private_data_len = ep->plen;
1312 event.private_data = ep->mpa_pkt +
1313 sizeof(struct mpa_message);
1317 pr_debug("%s ep %p tid %u status %d\n", __func__, ep,
1319 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1320 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1323 deref_cm_id(&ep->com);
1326 static int connect_request_upcall(struct c4iw_ep *ep)
1328 struct iw_cm_event event;
1331 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1332 memset(&event, 0, sizeof(event));
1333 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1334 memcpy(&event.local_addr, &ep->com.local_addr,
1335 sizeof(ep->com.local_addr));
1336 memcpy(&event.remote_addr, &ep->com.remote_addr,
1337 sizeof(ep->com.remote_addr));
1338 event.provider_data = ep;
1339 if (!ep->tried_with_mpa_v1) {
1340 /* this means MPA_v2 is used */
1341 event.ord = ep->ord;
1342 event.ird = ep->ird;
1343 event.private_data_len = ep->plen -
1344 sizeof(struct mpa_v2_conn_params);
1345 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1346 sizeof(struct mpa_v2_conn_params);
1348 /* this means MPA_v1 is used. Send max supported */
1349 event.ord = cur_max_read_depth(ep->com.dev);
1350 event.ird = cur_max_read_depth(ep->com.dev);
1351 event.private_data_len = ep->plen;
1352 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1354 c4iw_get_ep(&ep->com);
1355 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1358 c4iw_put_ep(&ep->com);
1359 set_bit(CONNREQ_UPCALL, &ep->com.history);
1360 c4iw_put_ep(&ep->parent_ep->com);
1364 static void established_upcall(struct c4iw_ep *ep)
1366 struct iw_cm_event event;
1368 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1369 memset(&event, 0, sizeof(event));
1370 event.event = IW_CM_EVENT_ESTABLISHED;
1371 event.ird = ep->ord;
1372 event.ord = ep->ird;
1373 if (ep->com.cm_id) {
1374 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1375 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1376 set_bit(ESTAB_UPCALL, &ep->com.history);
1380 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1382 struct sk_buff *skb;
1383 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1386 pr_debug("%s ep %p tid %u credits %u\n",
1387 __func__, ep, ep->hwtid, credits);
1388 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1390 pr_err("update_rx_credits - cannot alloc skb!\n");
1395 * If we couldn't specify the entire rcv window at connection setup
1396 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1397 * then add the overage in to the credits returned.
1399 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1400 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1402 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1403 RX_DACK_MODE_V(dack_mode);
1405 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1408 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1412 #define RELAXED_IRD_NEGOTIATION 1
1415 * process_mpa_reply - process streaming mode MPA reply
1419 * 0 upon success indicating a connect request was delivered to the ULP
1420 * or the mpa request is incomplete but valid so far.
1422 * 1 if a failure requires the caller to close the connection.
1424 * 2 if a failure requires the caller to abort the connection.
1426 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1428 struct mpa_message *mpa;
1429 struct mpa_v2_conn_params *mpa_v2_params;
1431 u16 resp_ird, resp_ord;
1432 u8 rtr_mismatch = 0, insuff_ird = 0;
1433 struct c4iw_qp_attributes attrs;
1434 enum c4iw_qp_attr_mask mask;
1438 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1441 * If we get more than the supported amount of private data
1442 * then we must fail this connection.
1444 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1446 goto err_stop_timer;
1450 * copy the new data into our accumulation buffer.
1452 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1454 ep->mpa_pkt_len += skb->len;
1457 * if we don't even have the mpa message, then bail.
1459 if (ep->mpa_pkt_len < sizeof(*mpa))
1461 mpa = (struct mpa_message *) ep->mpa_pkt;
1463 /* Validate MPA header. */
1464 if (mpa->revision > mpa_rev) {
1465 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1466 __func__, mpa_rev, mpa->revision);
1468 goto err_stop_timer;
1470 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1472 goto err_stop_timer;
1475 plen = ntohs(mpa->private_data_size);
1478 * Fail if there's too much private data.
1480 if (plen > MPA_MAX_PRIVATE_DATA) {
1482 goto err_stop_timer;
1486 * If plen does not account for pkt size
1488 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1490 goto err_stop_timer;
1493 ep->plen = (u8) plen;
1496 * If we don't have all the pdata yet, then bail.
1497 * We'll continue process when more data arrives.
1499 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1502 if (mpa->flags & MPA_REJECT) {
1503 err = -ECONNREFUSED;
1504 goto err_stop_timer;
1508 * Stop mpa timer. If it expired, then
1509 * we ignore the MPA reply. process_timeout()
1510 * will abort the connection.
1512 if (stop_ep_timer(ep))
1516 * If we get here we have accumulated the entire mpa
1517 * start reply message including private data. And
1518 * the MPA header is valid.
1520 __state_set(&ep->com, FPDU_MODE);
1521 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1522 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1523 ep->mpa_attr.version = mpa->revision;
1524 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1526 if (mpa->revision == 2) {
1527 ep->mpa_attr.enhanced_rdma_conn =
1528 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1529 if (ep->mpa_attr.enhanced_rdma_conn) {
1530 mpa_v2_params = (struct mpa_v2_conn_params *)
1531 (ep->mpa_pkt + sizeof(*mpa));
1532 resp_ird = ntohs(mpa_v2_params->ird) &
1533 MPA_V2_IRD_ORD_MASK;
1534 resp_ord = ntohs(mpa_v2_params->ord) &
1535 MPA_V2_IRD_ORD_MASK;
1536 pr_debug("%s responder ird %u ord %u ep ird %u ord %u\n",
1538 resp_ird, resp_ord, ep->ird, ep->ord);
1541 * This is a double-check. Ideally, below checks are
1542 * not required since ird/ord stuff has been taken
1543 * care of in c4iw_accept_cr
1545 if (ep->ird < resp_ord) {
1546 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1547 ep->com.dev->rdev.lldi.max_ordird_qp)
1551 } else if (ep->ird > resp_ord) {
1554 if (ep->ord > resp_ird) {
1555 if (RELAXED_IRD_NEGOTIATION)
1566 if (ntohs(mpa_v2_params->ird) &
1567 MPA_V2_PEER2PEER_MODEL) {
1568 if (ntohs(mpa_v2_params->ord) &
1569 MPA_V2_RDMA_WRITE_RTR)
1570 ep->mpa_attr.p2p_type =
1571 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1572 else if (ntohs(mpa_v2_params->ord) &
1573 MPA_V2_RDMA_READ_RTR)
1574 ep->mpa_attr.p2p_type =
1575 FW_RI_INIT_P2PTYPE_READ_REQ;
1578 } else if (mpa->revision == 1)
1580 ep->mpa_attr.p2p_type = p2p_type;
1582 pr_debug("%s - crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1583 __func__, ep->mpa_attr.crc_enabled,
1584 ep->mpa_attr.recv_marker_enabled,
1585 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1586 ep->mpa_attr.p2p_type, p2p_type);
1589 * If responder's RTR does not match with that of initiator, assign
1590 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1591 * generated when moving QP to RTS state.
1592 * A TERM message will be sent after QP has moved to RTS state
1594 if ((ep->mpa_attr.version == 2) && peer2peer &&
1595 (ep->mpa_attr.p2p_type != p2p_type)) {
1596 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1600 attrs.mpa_attr = ep->mpa_attr;
1601 attrs.max_ird = ep->ird;
1602 attrs.max_ord = ep->ord;
1603 attrs.llp_stream_handle = ep;
1604 attrs.next_state = C4IW_QP_STATE_RTS;
1606 mask = C4IW_QP_ATTR_NEXT_STATE |
1607 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1608 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1610 /* bind QP and TID with INIT_WR */
1611 err = c4iw_modify_qp(ep->com.qp->rhp,
1612 ep->com.qp, mask, &attrs, 1);
1617 * If responder's RTR requirement did not match with what initiator
1618 * supports, generate TERM message
1621 pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1622 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1623 attrs.ecode = MPA_NOMATCH_RTR;
1624 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1625 attrs.send_term = 1;
1626 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1627 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1634 * Generate TERM if initiator IRD is not sufficient for responder
1635 * provided ORD. Currently, we do the same behaviour even when
1636 * responder provided IRD is also not sufficient as regards to
1640 pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1641 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1642 attrs.ecode = MPA_INSUFF_IRD;
1643 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1644 attrs.send_term = 1;
1645 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1646 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1657 connect_reply_upcall(ep, err);
1662 * process_mpa_request - process streaming mode MPA request
1666 * 0 upon success indicating a connect request was delivered to the ULP
1667 * or the mpa request is incomplete but valid so far.
1669 * 1 if a failure requires the caller to close the connection.
1671 * 2 if a failure requires the caller to abort the connection.
1673 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1675 struct mpa_message *mpa;
1676 struct mpa_v2_conn_params *mpa_v2_params;
1679 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1682 * If we get more than the supported amount of private data
1683 * then we must fail this connection.
1685 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1686 goto err_stop_timer;
1688 pr_debug("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1691 * Copy the new data into our accumulation buffer.
1693 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1695 ep->mpa_pkt_len += skb->len;
1698 * If we don't even have the mpa message, then bail.
1699 * We'll continue process when more data arrives.
1701 if (ep->mpa_pkt_len < sizeof(*mpa))
1704 pr_debug("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1705 mpa = (struct mpa_message *) ep->mpa_pkt;
1708 * Validate MPA Header.
1710 if (mpa->revision > mpa_rev) {
1711 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1712 __func__, mpa_rev, mpa->revision);
1713 goto err_stop_timer;
1716 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1717 goto err_stop_timer;
1719 plen = ntohs(mpa->private_data_size);
1722 * Fail if there's too much private data.
1724 if (plen > MPA_MAX_PRIVATE_DATA)
1725 goto err_stop_timer;
1728 * If plen does not account for pkt size
1730 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1731 goto err_stop_timer;
1732 ep->plen = (u8) plen;
1735 * If we don't have all the pdata yet, then bail.
1737 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1741 * If we get here we have accumulated the entire mpa
1742 * start reply message including private data.
1744 ep->mpa_attr.initiator = 0;
1745 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1746 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1747 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1748 ep->mpa_attr.version = mpa->revision;
1749 if (mpa->revision == 1)
1750 ep->tried_with_mpa_v1 = 1;
1751 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1753 if (mpa->revision == 2) {
1754 ep->mpa_attr.enhanced_rdma_conn =
1755 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1756 if (ep->mpa_attr.enhanced_rdma_conn) {
1757 mpa_v2_params = (struct mpa_v2_conn_params *)
1758 (ep->mpa_pkt + sizeof(*mpa));
1759 ep->ird = ntohs(mpa_v2_params->ird) &
1760 MPA_V2_IRD_ORD_MASK;
1761 ep->ird = min_t(u32, ep->ird,
1762 cur_max_read_depth(ep->com.dev));
1763 ep->ord = ntohs(mpa_v2_params->ord) &
1764 MPA_V2_IRD_ORD_MASK;
1765 ep->ord = min_t(u32, ep->ord,
1766 cur_max_read_depth(ep->com.dev));
1767 pr_debug("%s initiator ird %u ord %u\n",
1768 __func__, ep->ird, ep->ord);
1769 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1771 if (ntohs(mpa_v2_params->ord) &
1772 MPA_V2_RDMA_WRITE_RTR)
1773 ep->mpa_attr.p2p_type =
1774 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1775 else if (ntohs(mpa_v2_params->ord) &
1776 MPA_V2_RDMA_READ_RTR)
1777 ep->mpa_attr.p2p_type =
1778 FW_RI_INIT_P2PTYPE_READ_REQ;
1781 } else if (mpa->revision == 1)
1783 ep->mpa_attr.p2p_type = p2p_type;
1785 pr_debug("%s - crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1787 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1788 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1789 ep->mpa_attr.p2p_type);
1791 __state_set(&ep->com, MPA_REQ_RCVD);
1794 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1795 if (ep->parent_ep->com.state != DEAD) {
1796 if (connect_request_upcall(ep))
1797 goto err_unlock_parent;
1799 goto err_unlock_parent;
1801 mutex_unlock(&ep->parent_ep->com.mutex);
1805 mutex_unlock(&ep->parent_ep->com.mutex);
1808 (void)stop_ep_timer(ep);
1813 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1816 struct cpl_rx_data *hdr = cplhdr(skb);
1817 unsigned int dlen = ntohs(hdr->len);
1818 unsigned int tid = GET_TID(hdr);
1819 __u8 status = hdr->status;
1822 ep = get_ep_from_tid(dev, tid);
1825 pr_debug("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1826 skb_pull(skb, sizeof(*hdr));
1827 skb_trim(skb, dlen);
1828 mutex_lock(&ep->com.mutex);
1830 switch (ep->com.state) {
1832 update_rx_credits(ep, dlen);
1833 ep->rcv_seq += dlen;
1834 disconnect = process_mpa_reply(ep, skb);
1837 update_rx_credits(ep, dlen);
1838 ep->rcv_seq += dlen;
1839 disconnect = process_mpa_request(ep, skb);
1842 struct c4iw_qp_attributes attrs;
1844 update_rx_credits(ep, dlen);
1845 BUG_ON(!ep->com.qp);
1847 pr_err("%s Unexpected streaming data." \
1848 " qpid %u ep %p state %d tid %u status %d\n",
1849 __func__, ep->com.qp->wq.sq.qid, ep,
1850 ep->com.state, ep->hwtid, status);
1851 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1852 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1853 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1860 mutex_unlock(&ep->com.mutex);
1862 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1863 c4iw_put_ep(&ep->com);
1867 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1870 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1872 unsigned int tid = GET_TID(rpl);
1874 ep = get_ep_from_tid(dev, tid);
1876 pr_warn("Abort rpl to freed endpoint\n");
1879 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1880 mutex_lock(&ep->com.mutex);
1881 switch (ep->com.state) {
1883 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1884 __state_set(&ep->com, DEAD);
1888 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1891 mutex_unlock(&ep->com.mutex);
1894 release_ep_resources(ep);
1895 c4iw_put_ep(&ep->com);
1899 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1901 struct sk_buff *skb;
1902 struct fw_ofld_connection_wr *req;
1903 unsigned int mtu_idx;
1905 struct sockaddr_in *sin;
1908 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1909 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1910 memset(req, 0, sizeof(*req));
1911 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1912 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1913 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1914 ep->com.dev->rdev.lldi.ports[0],
1916 sin = (struct sockaddr_in *)&ep->com.local_addr;
1917 req->le.lport = sin->sin_port;
1918 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1919 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1920 req->le.pport = sin->sin_port;
1921 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1922 req->tcb.t_state_to_astid =
1923 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1924 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1925 req->tcb.cplrxdataack_cplpassacceptrpl =
1926 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1927 req->tcb.tx_max = (__force __be32) jiffies;
1928 req->tcb.rcv_adv = htons(1);
1929 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1930 enable_tcp_timestamps,
1931 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1932 wscale = cxgb_compute_wscale(rcv_win);
1935 * Specify the largest window that will fit in opt0. The
1936 * remainder will be specified in the rx_data_ack.
1938 win = ep->rcv_win >> 10;
1939 if (win > RCV_BUFSIZ_M)
1942 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1943 (nocong ? NO_CONG_F : 0) |
1946 WND_SCALE_V(wscale) |
1947 MSS_IDX_V(mtu_idx) |
1948 L2T_IDX_V(ep->l2t->idx) |
1949 TX_CHAN_V(ep->tx_chan) |
1950 SMAC_SEL_V(ep->smac_idx) |
1951 DSCP_V(ep->tos >> 2) |
1952 ULP_MODE_V(ULP_MODE_TCPDDP) |
1954 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1955 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1957 CCTRL_ECN_V(enable_ecn) |
1958 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1959 if (enable_tcp_timestamps)
1960 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1961 if (enable_tcp_sack)
1962 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1963 if (wscale && enable_tcp_window_scaling)
1964 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1965 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1966 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1967 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1968 set_bit(ACT_OFLD_CONN, &ep->com.history);
1969 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1973 * Some of the error codes above implicitly indicate that there is no TID
1974 * allocated with the result of an ACT_OPEN. We use this predicate to make
1977 static inline int act_open_has_tid(int status)
1979 return (status != CPL_ERR_TCAM_PARITY &&
1980 status != CPL_ERR_TCAM_MISS &&
1981 status != CPL_ERR_TCAM_FULL &&
1982 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1983 status != CPL_ERR_CONN_EXIST);
1986 static char *neg_adv_str(unsigned int status)
1989 case CPL_ERR_RTX_NEG_ADVICE:
1990 return "Retransmit timeout";
1991 case CPL_ERR_PERSIST_NEG_ADVICE:
1992 return "Persist timeout";
1993 case CPL_ERR_KEEPALV_NEG_ADVICE:
1994 return "Keepalive timeout";
2000 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2002 ep->snd_win = snd_win;
2003 ep->rcv_win = rcv_win;
2004 pr_debug("%s snd_win %d rcv_win %d\n",
2005 __func__, ep->snd_win, ep->rcv_win);
2008 #define ACT_OPEN_RETRY_COUNT 2
2010 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2011 struct dst_entry *dst, struct c4iw_dev *cdev,
2012 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2014 struct neighbour *n;
2016 struct net_device *pdev;
2018 n = dst_neigh_lookup(dst, peer_ip);
2024 if (n->dev->flags & IFF_LOOPBACK) {
2026 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2027 else if (IS_ENABLED(CONFIG_IPV6))
2028 for_each_netdev(&init_net, pdev) {
2029 if (ipv6_chk_addr(&init_net,
2030 (struct in6_addr *)peer_ip,
2041 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2042 n, pdev, rt_tos2priority(tos));
2047 ep->mtu = pdev->mtu;
2048 ep->tx_chan = cxgb4_port_chan(pdev);
2049 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2050 cxgb4_port_viid(pdev));
2051 step = cdev->rdev.lldi.ntxq /
2052 cdev->rdev.lldi.nchan;
2053 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2054 step = cdev->rdev.lldi.nrxq /
2055 cdev->rdev.lldi.nchan;
2056 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2057 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2058 cxgb4_port_idx(pdev) * step];
2059 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2062 pdev = get_real_dev(n->dev);
2063 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2067 ep->mtu = dst_mtu(dst);
2068 ep->tx_chan = cxgb4_port_chan(pdev);
2069 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2070 cxgb4_port_viid(pdev));
2071 step = cdev->rdev.lldi.ntxq /
2072 cdev->rdev.lldi.nchan;
2073 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2074 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2075 step = cdev->rdev.lldi.nrxq /
2076 cdev->rdev.lldi.nchan;
2077 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2078 cxgb4_port_idx(pdev) * step];
2079 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2082 ep->retry_with_mpa_v1 = 0;
2083 ep->tried_with_mpa_v1 = 0;
2095 static int c4iw_reconnect(struct c4iw_ep *ep)
2099 struct sockaddr_in *laddr = (struct sockaddr_in *)
2100 &ep->com.cm_id->m_local_addr;
2101 struct sockaddr_in *raddr = (struct sockaddr_in *)
2102 &ep->com.cm_id->m_remote_addr;
2103 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2104 &ep->com.cm_id->m_local_addr;
2105 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2106 &ep->com.cm_id->m_remote_addr;
2110 pr_debug("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2111 init_timer(&ep->timer);
2112 c4iw_init_wr_wait(&ep->com.wr_wait);
2114 /* When MPA revision is different on nodes, the node with MPA_rev=2
2115 * tries to reconnect with MPA_rev 1 for the same EP through
2116 * c4iw_reconnect(), where the same EP is assigned with new tid for
2117 * further connection establishment. As we are using the same EP pointer
2118 * for reconnect, few skbs are used during the previous c4iw_connect(),
2119 * which leaves the EP with inadequate skbs for further
2120 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2121 * skb_list() during peer_abort(). Allocate skbs which is already used.
2123 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2124 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2130 * Allocate an active TID to initiate a TCP connection.
2132 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2133 if (ep->atid == -1) {
2134 pr_err("%s - cannot alloc atid\n", __func__);
2138 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2141 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2142 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2143 laddr->sin_addr.s_addr,
2144 raddr->sin_addr.s_addr,
2146 raddr->sin_port, ep->com.cm_id->tos);
2148 ra = (__u8 *)&raddr->sin_addr;
2150 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2152 laddr6->sin6_addr.s6_addr,
2153 raddr6->sin6_addr.s6_addr,
2155 raddr6->sin6_port, 0,
2156 raddr6->sin6_scope_id);
2158 ra = (__u8 *)&raddr6->sin6_addr;
2161 pr_err("%s - cannot find route\n", __func__);
2162 err = -EHOSTUNREACH;
2165 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2166 ep->com.dev->rdev.lldi.adapter_type,
2167 ep->com.cm_id->tos);
2169 pr_err("%s - cannot alloc l2e\n", __func__);
2173 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2174 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2177 state_set(&ep->com, CONNECTING);
2178 ep->tos = ep->com.cm_id->tos;
2180 /* send connect request to rnic */
2181 err = send_connect(ep);
2185 cxgb4_l2t_release(ep->l2t);
2187 dst_release(ep->dst);
2189 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2190 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2193 * remember to send notification to upper layer.
2194 * We are in here so the upper layer is not aware that this is
2195 * re-connect attempt and so, upper layer is still waiting for
2196 * response of 1st connect request.
2198 connect_reply_upcall(ep, -ECONNRESET);
2200 c4iw_put_ep(&ep->com);
2205 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2208 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2209 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2210 ntohl(rpl->atid_status)));
2211 struct tid_info *t = dev->rdev.lldi.tids;
2212 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2213 struct sockaddr_in *la;
2214 struct sockaddr_in *ra;
2215 struct sockaddr_in6 *la6;
2216 struct sockaddr_in6 *ra6;
2219 ep = lookup_atid(t, atid);
2220 la = (struct sockaddr_in *)&ep->com.local_addr;
2221 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2222 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2223 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2225 pr_debug("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2226 status, status2errno(status));
2228 if (cxgb_is_neg_adv(status)) {
2229 pr_debug("%s Connection problems for atid %u status %u (%s)\n",
2230 __func__, atid, status, neg_adv_str(status));
2231 ep->stats.connect_neg_adv++;
2232 mutex_lock(&dev->rdev.stats.lock);
2233 dev->rdev.stats.neg_adv++;
2234 mutex_unlock(&dev->rdev.stats.lock);
2238 set_bit(ACT_OPEN_RPL, &ep->com.history);
2241 * Log interesting failures.
2244 case CPL_ERR_CONN_RESET:
2245 case CPL_ERR_CONN_TIMEDOUT:
2247 case CPL_ERR_TCAM_FULL:
2248 mutex_lock(&dev->rdev.stats.lock);
2249 dev->rdev.stats.tcam_full++;
2250 mutex_unlock(&dev->rdev.stats.lock);
2251 if (ep->com.local_addr.ss_family == AF_INET &&
2252 dev->rdev.lldi.enable_fw_ofld_conn) {
2253 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2254 ntohl(rpl->atid_status))));
2260 case CPL_ERR_CONN_EXIST:
2261 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2262 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2263 if (ep->com.remote_addr.ss_family == AF_INET6) {
2264 struct sockaddr_in6 *sin6 =
2265 (struct sockaddr_in6 *)
2266 &ep->com.local_addr;
2268 ep->com.dev->rdev.lldi.ports[0],
2270 &sin6->sin6_addr.s6_addr, 1);
2272 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2274 cxgb4_free_atid(t, atid);
2275 dst_release(ep->dst);
2276 cxgb4_l2t_release(ep->l2t);
2282 if (ep->com.local_addr.ss_family == AF_INET) {
2283 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2284 atid, status, status2errno(status),
2285 &la->sin_addr.s_addr, ntohs(la->sin_port),
2286 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2288 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2289 atid, status, status2errno(status),
2290 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2291 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2297 connect_reply_upcall(ep, status2errno(status));
2298 state_set(&ep->com, DEAD);
2300 if (ep->com.remote_addr.ss_family == AF_INET6) {
2301 struct sockaddr_in6 *sin6 =
2302 (struct sockaddr_in6 *)&ep->com.local_addr;
2303 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2304 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2306 if (status && act_open_has_tid(status))
2307 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2309 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2310 cxgb4_free_atid(t, atid);
2311 dst_release(ep->dst);
2312 cxgb4_l2t_release(ep->l2t);
2313 c4iw_put_ep(&ep->com);
2318 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2320 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2321 unsigned int stid = GET_TID(rpl);
2322 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2325 pr_debug("%s stid %d lookup failure!\n", __func__, stid);
2328 pr_debug("%s ep %p status %d error %d\n", __func__, ep,
2329 rpl->status, status2errno(rpl->status));
2330 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2331 c4iw_put_ep(&ep->com);
2336 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2338 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2339 unsigned int stid = GET_TID(rpl);
2340 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2342 pr_debug("%s ep %p\n", __func__, ep);
2343 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2344 c4iw_put_ep(&ep->com);
2348 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2349 struct cpl_pass_accept_req *req)
2351 struct cpl_pass_accept_rpl *rpl;
2352 unsigned int mtu_idx;
2356 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2358 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2360 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2361 BUG_ON(skb_cloned(skb));
2365 if (!is_t4(adapter_type)) {
2366 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2368 INIT_TP_WR(rpl5, ep->hwtid);
2370 skb_trim(skb, sizeof(*rpl));
2371 INIT_TP_WR(rpl, ep->hwtid);
2373 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2376 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2377 enable_tcp_timestamps && req->tcpopt.tstamp,
2378 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2379 wscale = cxgb_compute_wscale(rcv_win);
2382 * Specify the largest window that will fit in opt0. The
2383 * remainder will be specified in the rx_data_ack.
2385 win = ep->rcv_win >> 10;
2386 if (win > RCV_BUFSIZ_M)
2388 opt0 = (nocong ? NO_CONG_F : 0) |
2391 WND_SCALE_V(wscale) |
2392 MSS_IDX_V(mtu_idx) |
2393 L2T_IDX_V(ep->l2t->idx) |
2394 TX_CHAN_V(ep->tx_chan) |
2395 SMAC_SEL_V(ep->smac_idx) |
2396 DSCP_V(ep->tos >> 2) |
2397 ULP_MODE_V(ULP_MODE_TCPDDP) |
2399 opt2 = RX_CHANNEL_V(0) |
2400 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2402 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2403 opt2 |= TSTAMPS_EN_F;
2404 if (enable_tcp_sack && req->tcpopt.sack)
2406 if (wscale && enable_tcp_window_scaling)
2407 opt2 |= WND_SCALE_EN_F;
2409 const struct tcphdr *tcph;
2410 u32 hlen = ntohl(req->hdr_len);
2412 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2413 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2416 tcph = (const void *)(req + 1) +
2417 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2418 if (tcph->ece && tcph->cwr)
2419 opt2 |= CCTRL_ECN_V(1);
2421 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2422 u32 isn = (prandom_u32() & ~7UL) - 1;
2423 opt2 |= T5_OPT_2_VALID_F;
2424 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2427 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2430 rpl5->iss = cpu_to_be32(isn);
2431 pr_debug("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2434 rpl->opt0 = cpu_to_be64(opt0);
2435 rpl->opt2 = cpu_to_be32(opt2);
2436 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2437 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2439 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2442 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2444 pr_debug("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2445 BUG_ON(skb_cloned(skb));
2446 skb_trim(skb, sizeof(struct cpl_tid_release));
2447 release_tid(&dev->rdev, hwtid, skb);
2451 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2453 struct c4iw_ep *child_ep = NULL, *parent_ep;
2454 struct cpl_pass_accept_req *req = cplhdr(skb);
2455 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2456 struct tid_info *t = dev->rdev.lldi.tids;
2457 unsigned int hwtid = GET_TID(req);
2458 struct dst_entry *dst;
2459 __u8 local_ip[16], peer_ip[16];
2460 __be16 local_port, peer_port;
2461 struct sockaddr_in6 *sin6;
2463 u16 peer_mss = ntohs(req->tcpopt.mss);
2465 unsigned short hdrs;
2466 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2468 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2470 pr_debug("%s connect request on invalid stid %d\n",
2475 if (state_read(&parent_ep->com) != LISTEN) {
2476 pr_debug("%s - listening ep not in LISTEN\n", __func__);
2480 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2481 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2483 /* Find output route */
2485 pr_debug("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 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_route(&dev->rdev.lldi, get_real_dev,
2490 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2491 local_port, peer_port, tos);
2493 pr_debug("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2494 , __func__, parent_ep, hwtid,
2495 local_ip, peer_ip, ntohs(local_port),
2496 ntohs(peer_port), peer_mss);
2497 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2498 local_ip, peer_ip, local_port, peer_port,
2499 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2500 ((struct sockaddr_in6 *)
2501 &parent_ep->com.local_addr)->sin6_scope_id);
2504 pr_err("%s - failed to find dst entry!\n", __func__);
2508 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2510 pr_err("%s - failed to allocate ep entry!\n", __func__);
2515 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2516 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2518 pr_err("%s - failed to allocate l2t entry!\n", __func__);
2524 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2525 sizeof(struct tcphdr) +
2526 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2527 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2528 child_ep->mtu = peer_mss + hdrs;
2530 skb_queue_head_init(&child_ep->com.ep_skb_list);
2531 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2534 state_set(&child_ep->com, CONNECTING);
2535 child_ep->com.dev = dev;
2536 child_ep->com.cm_id = NULL;
2539 struct sockaddr_in *sin = (struct sockaddr_in *)
2540 &child_ep->com.local_addr;
2542 sin->sin_family = AF_INET;
2543 sin->sin_port = local_port;
2544 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2546 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2547 sin->sin_family = AF_INET;
2548 sin->sin_port = ((struct sockaddr_in *)
2549 &parent_ep->com.local_addr)->sin_port;
2550 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2552 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2553 sin->sin_family = AF_INET;
2554 sin->sin_port = peer_port;
2555 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2557 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2558 sin6->sin6_family = PF_INET6;
2559 sin6->sin6_port = local_port;
2560 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2562 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2563 sin6->sin6_family = PF_INET6;
2564 sin6->sin6_port = ((struct sockaddr_in6 *)
2565 &parent_ep->com.local_addr)->sin6_port;
2566 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2568 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2569 sin6->sin6_family = PF_INET6;
2570 sin6->sin6_port = peer_port;
2571 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2574 c4iw_get_ep(&parent_ep->com);
2575 child_ep->parent_ep = parent_ep;
2576 child_ep->tos = tos;
2577 child_ep->dst = dst;
2578 child_ep->hwtid = hwtid;
2580 pr_debug("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2581 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2583 init_timer(&child_ep->timer);
2584 cxgb4_insert_tid(t, child_ep, hwtid);
2585 insert_ep_tid(child_ep);
2586 if (accept_cr(child_ep, skb, req)) {
2587 c4iw_put_ep(&parent_ep->com);
2588 release_ep_resources(child_ep);
2590 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2593 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2594 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2595 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2599 c4iw_put_ep(&child_ep->com);
2601 reject_cr(dev, hwtid, skb);
2603 c4iw_put_ep(&parent_ep->com);
2608 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2611 struct cpl_pass_establish *req = cplhdr(skb);
2612 unsigned int tid = GET_TID(req);
2615 ep = get_ep_from_tid(dev, tid);
2616 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2617 ep->snd_seq = be32_to_cpu(req->snd_isn);
2618 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2620 pr_debug("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2621 ntohs(req->tcp_opt));
2623 set_emss(ep, ntohs(req->tcp_opt));
2625 dst_confirm(ep->dst);
2626 mutex_lock(&ep->com.mutex);
2627 ep->com.state = MPA_REQ_WAIT;
2629 set_bit(PASS_ESTAB, &ep->com.history);
2630 ret = send_flowc(ep);
2631 mutex_unlock(&ep->com.mutex);
2633 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2634 c4iw_put_ep(&ep->com);
2639 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2641 struct cpl_peer_close *hdr = cplhdr(skb);
2643 struct c4iw_qp_attributes attrs;
2646 unsigned int tid = GET_TID(hdr);
2649 ep = get_ep_from_tid(dev, tid);
2653 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2654 dst_confirm(ep->dst);
2656 set_bit(PEER_CLOSE, &ep->com.history);
2657 mutex_lock(&ep->com.mutex);
2658 switch (ep->com.state) {
2660 __state_set(&ep->com, CLOSING);
2663 __state_set(&ep->com, CLOSING);
2664 connect_reply_upcall(ep, -ECONNRESET);
2669 * We're gonna mark this puppy DEAD, but keep
2670 * the reference on it until the ULP accepts or
2671 * rejects the CR. Also wake up anyone waiting
2672 * in rdma connection migration (see c4iw_accept_cr()).
2674 __state_set(&ep->com, CLOSING);
2675 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2676 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2679 __state_set(&ep->com, CLOSING);
2680 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2681 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2685 __state_set(&ep->com, CLOSING);
2686 attrs.next_state = C4IW_QP_STATE_CLOSING;
2687 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2688 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2689 if (ret != -ECONNRESET) {
2690 peer_close_upcall(ep);
2698 __state_set(&ep->com, MORIBUND);
2702 (void)stop_ep_timer(ep);
2703 if (ep->com.cm_id && ep->com.qp) {
2704 attrs.next_state = C4IW_QP_STATE_IDLE;
2705 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2706 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2708 close_complete_upcall(ep, 0);
2709 __state_set(&ep->com, DEAD);
2719 mutex_unlock(&ep->com.mutex);
2721 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2723 release_ep_resources(ep);
2724 c4iw_put_ep(&ep->com);
2728 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2730 struct cpl_abort_req_rss *req = cplhdr(skb);
2732 struct sk_buff *rpl_skb;
2733 struct c4iw_qp_attributes attrs;
2736 unsigned int tid = GET_TID(req);
2737 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2739 ep = get_ep_from_tid(dev, tid);
2743 if (cxgb_is_neg_adv(req->status)) {
2744 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n",
2745 __func__, ep->hwtid, req->status,
2746 neg_adv_str(req->status));
2747 ep->stats.abort_neg_adv++;
2748 mutex_lock(&dev->rdev.stats.lock);
2749 dev->rdev.stats.neg_adv++;
2750 mutex_unlock(&dev->rdev.stats.lock);
2753 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2755 set_bit(PEER_ABORT, &ep->com.history);
2758 * Wake up any threads in rdma_init() or rdma_fini().
2759 * However, this is not needed if com state is just
2762 if (ep->com.state != MPA_REQ_SENT)
2763 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2765 mutex_lock(&ep->com.mutex);
2766 switch (ep->com.state) {
2768 c4iw_put_ep(&ep->parent_ep->com);
2771 (void)stop_ep_timer(ep);
2774 (void)stop_ep_timer(ep);
2775 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2776 connect_reply_upcall(ep, -ECONNRESET);
2779 * we just don't send notification upwards because we
2780 * want to retry with mpa_v1 without upper layers even
2783 * do some housekeeping so as to re-initiate the
2786 pr_debug("%s: mpa_rev=%d. Retrying with mpav1\n",
2788 ep->retry_with_mpa_v1 = 1;
2800 if (ep->com.cm_id && ep->com.qp) {
2801 attrs.next_state = C4IW_QP_STATE_ERROR;
2802 ret = c4iw_modify_qp(ep->com.qp->rhp,
2803 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2806 pr_err("%s - qp <- error failed!\n", __func__);
2808 peer_abort_upcall(ep);
2813 pr_debug("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2814 mutex_unlock(&ep->com.mutex);
2820 dst_confirm(ep->dst);
2821 if (ep->com.state != ABORTING) {
2822 __state_set(&ep->com, DEAD);
2823 /* we don't release if we want to retry with mpa_v1 */
2824 if (!ep->retry_with_mpa_v1)
2827 mutex_unlock(&ep->com.mutex);
2829 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2830 if (WARN_ON(!rpl_skb)) {
2835 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2837 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2840 release_ep_resources(ep);
2841 else if (ep->retry_with_mpa_v1) {
2842 if (ep->com.remote_addr.ss_family == AF_INET6) {
2843 struct sockaddr_in6 *sin6 =
2844 (struct sockaddr_in6 *)
2845 &ep->com.local_addr;
2847 ep->com.dev->rdev.lldi.ports[0],
2848 (const u32 *)&sin6->sin6_addr.s6_addr,
2851 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2852 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2853 dst_release(ep->dst);
2854 cxgb4_l2t_release(ep->l2t);
2859 c4iw_put_ep(&ep->com);
2860 /* Dereferencing ep, referenced in peer_abort_intr() */
2861 c4iw_put_ep(&ep->com);
2865 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2868 struct c4iw_qp_attributes attrs;
2869 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2871 unsigned int tid = GET_TID(rpl);
2873 ep = get_ep_from_tid(dev, tid);
2877 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2880 /* The cm_id may be null if we failed to connect */
2881 mutex_lock(&ep->com.mutex);
2882 set_bit(CLOSE_CON_RPL, &ep->com.history);
2883 switch (ep->com.state) {
2885 __state_set(&ep->com, MORIBUND);
2888 (void)stop_ep_timer(ep);
2889 if ((ep->com.cm_id) && (ep->com.qp)) {
2890 attrs.next_state = C4IW_QP_STATE_IDLE;
2891 c4iw_modify_qp(ep->com.qp->rhp,
2893 C4IW_QP_ATTR_NEXT_STATE,
2896 close_complete_upcall(ep, 0);
2897 __state_set(&ep->com, DEAD);
2907 mutex_unlock(&ep->com.mutex);
2909 release_ep_resources(ep);
2910 c4iw_put_ep(&ep->com);
2914 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2916 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2917 unsigned int tid = GET_TID(rpl);
2919 struct c4iw_qp_attributes attrs;
2921 ep = get_ep_from_tid(dev, tid);
2924 if (ep && ep->com.qp) {
2925 pr_warn("TERM received tid %u qpid %u\n",
2926 tid, ep->com.qp->wq.sq.qid);
2927 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2928 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2929 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2931 pr_warn("TERM received tid %u no ep/qp\n", tid);
2932 c4iw_put_ep(&ep->com);
2938 * Upcall from the adapter indicating data has been transmitted.
2939 * For us its just the single MPA request or reply. We can now free
2940 * the skb holding the mpa message.
2942 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2945 struct cpl_fw4_ack *hdr = cplhdr(skb);
2946 u8 credits = hdr->credits;
2947 unsigned int tid = GET_TID(hdr);
2950 ep = get_ep_from_tid(dev, tid);
2953 pr_debug("%s ep %p tid %u credits %u\n",
2954 __func__, ep, ep->hwtid, credits);
2956 pr_debug("%s 0 credit ack ep %p tid %u state %u\n",
2957 __func__, ep, ep->hwtid, state_read(&ep->com));
2961 dst_confirm(ep->dst);
2963 pr_debug("%s last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
2964 __func__, ep, ep->hwtid,
2965 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2966 mutex_lock(&ep->com.mutex);
2967 kfree_skb(ep->mpa_skb);
2969 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2971 mutex_unlock(&ep->com.mutex);
2974 c4iw_put_ep(&ep->com);
2978 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2981 struct c4iw_ep *ep = to_ep(cm_id);
2983 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2985 mutex_lock(&ep->com.mutex);
2986 if (ep->com.state != MPA_REQ_RCVD) {
2987 mutex_unlock(&ep->com.mutex);
2988 c4iw_put_ep(&ep->com);
2991 set_bit(ULP_REJECT, &ep->com.history);
2995 abort = send_mpa_reject(ep, pdata, pdata_len);
2996 mutex_unlock(&ep->com.mutex);
2999 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3000 c4iw_put_ep(&ep->com);
3004 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3007 struct c4iw_qp_attributes attrs;
3008 enum c4iw_qp_attr_mask mask;
3009 struct c4iw_ep *ep = to_ep(cm_id);
3010 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3011 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3014 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3016 mutex_lock(&ep->com.mutex);
3017 if (ep->com.state != MPA_REQ_RCVD) {
3024 set_bit(ULP_ACCEPT, &ep->com.history);
3025 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3026 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3031 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3032 if (conn_param->ord > ep->ird) {
3033 if (RELAXED_IRD_NEGOTIATION) {
3034 conn_param->ord = ep->ird;
3036 ep->ird = conn_param->ird;
3037 ep->ord = conn_param->ord;
3038 send_mpa_reject(ep, conn_param->private_data,
3039 conn_param->private_data_len);
3044 if (conn_param->ird < ep->ord) {
3045 if (RELAXED_IRD_NEGOTIATION &&
3046 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3047 conn_param->ird = ep->ord;
3054 ep->ird = conn_param->ird;
3055 ep->ord = conn_param->ord;
3057 if (ep->mpa_attr.version == 1) {
3058 if (peer2peer && ep->ird == 0)
3062 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3063 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3067 pr_debug("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3069 ep->com.cm_id = cm_id;
3070 ref_cm_id(&ep->com);
3074 /* bind QP to EP and move to RTS */
3075 attrs.mpa_attr = ep->mpa_attr;
3076 attrs.max_ird = ep->ird;
3077 attrs.max_ord = ep->ord;
3078 attrs.llp_stream_handle = ep;
3079 attrs.next_state = C4IW_QP_STATE_RTS;
3081 /* bind QP and TID with INIT_WR */
3082 mask = C4IW_QP_ATTR_NEXT_STATE |
3083 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3084 C4IW_QP_ATTR_MPA_ATTR |
3085 C4IW_QP_ATTR_MAX_IRD |
3086 C4IW_QP_ATTR_MAX_ORD;
3088 err = c4iw_modify_qp(ep->com.qp->rhp,
3089 ep->com.qp, mask, &attrs, 1);
3091 goto err_deref_cm_id;
3093 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3094 err = send_mpa_reply(ep, conn_param->private_data,
3095 conn_param->private_data_len);
3097 goto err_deref_cm_id;
3099 __state_set(&ep->com, FPDU_MODE);
3100 established_upcall(ep);
3101 mutex_unlock(&ep->com.mutex);
3102 c4iw_put_ep(&ep->com);
3105 deref_cm_id(&ep->com);
3109 mutex_unlock(&ep->com.mutex);
3111 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3112 c4iw_put_ep(&ep->com);
3116 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3118 struct in_device *ind;
3120 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3121 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3123 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3125 return -EADDRNOTAVAIL;
3126 for_primary_ifa(ind) {
3127 laddr->sin_addr.s_addr = ifa->ifa_address;
3128 raddr->sin_addr.s_addr = ifa->ifa_address;
3134 return found ? 0 : -EADDRNOTAVAIL;
3137 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3138 unsigned char banned_flags)
3140 struct inet6_dev *idev;
3141 int err = -EADDRNOTAVAIL;
3144 idev = __in6_dev_get(dev);
3146 struct inet6_ifaddr *ifp;
3148 read_lock_bh(&idev->lock);
3149 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3150 if (ifp->scope == IFA_LINK &&
3151 !(ifp->flags & banned_flags)) {
3152 memcpy(addr, &ifp->addr, 16);
3157 read_unlock_bh(&idev->lock);
3163 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3165 struct in6_addr uninitialized_var(addr);
3166 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3167 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3169 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3170 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3171 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3174 return -EADDRNOTAVAIL;
3177 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3179 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3182 struct sockaddr_in *laddr;
3183 struct sockaddr_in *raddr;
3184 struct sockaddr_in6 *laddr6;
3185 struct sockaddr_in6 *raddr6;
3189 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3190 (conn_param->ird > cur_max_read_depth(dev))) {
3194 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3196 pr_err("%s - cannot alloc ep\n", __func__);
3201 skb_queue_head_init(&ep->com.ep_skb_list);
3202 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3207 init_timer(&ep->timer);
3208 ep->plen = conn_param->private_data_len;
3210 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3211 conn_param->private_data, ep->plen);
3212 ep->ird = conn_param->ird;
3213 ep->ord = conn_param->ord;
3215 if (peer2peer && ep->ord == 0)
3218 ep->com.cm_id = cm_id;
3219 ref_cm_id(&ep->com);
3221 ep->com.qp = get_qhp(dev, conn_param->qpn);
3223 pr_debug("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3228 pr_debug("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3232 * Allocate an active TID to initiate a TCP connection.
3234 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3235 if (ep->atid == -1) {
3236 pr_err("%s - cannot alloc atid\n", __func__);
3240 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3242 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3243 sizeof(ep->com.local_addr));
3244 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3245 sizeof(ep->com.remote_addr));
3247 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3248 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3249 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3250 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3252 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3254 ra = (__u8 *)&raddr->sin_addr;
3257 * Handle loopback requests to INADDR_ANY.
3259 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3260 err = pick_local_ipaddrs(dev, cm_id);
3266 pr_debug("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3267 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3268 ra, ntohs(raddr->sin_port));
3269 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3270 laddr->sin_addr.s_addr,
3271 raddr->sin_addr.s_addr,
3273 raddr->sin_port, cm_id->tos);
3276 ra = (__u8 *)&raddr6->sin6_addr;
3279 * Handle loopback requests to INADDR_ANY.
3281 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3282 err = pick_local_ip6addrs(dev, cm_id);
3288 pr_debug("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3289 __func__, laddr6->sin6_addr.s6_addr,
3290 ntohs(laddr6->sin6_port),
3291 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3292 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3293 laddr6->sin6_addr.s6_addr,
3294 raddr6->sin6_addr.s6_addr,
3296 raddr6->sin6_port, 0,
3297 raddr6->sin6_scope_id);
3300 pr_err("%s - cannot find route\n", __func__);
3301 err = -EHOSTUNREACH;
3305 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3306 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3308 pr_err("%s - cannot alloc l2e\n", __func__);
3312 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3313 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3316 state_set(&ep->com, CONNECTING);
3317 ep->tos = cm_id->tos;
3319 /* send connect request to rnic */
3320 err = send_connect(ep);
3324 cxgb4_l2t_release(ep->l2t);
3326 dst_release(ep->dst);
3328 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3329 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3331 skb_queue_purge(&ep->com.ep_skb_list);
3332 deref_cm_id(&ep->com);
3334 c4iw_put_ep(&ep->com);
3339 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3342 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3343 &ep->com.local_addr;
3345 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3346 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3347 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3351 c4iw_init_wr_wait(&ep->com.wr_wait);
3352 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3353 ep->stid, &sin6->sin6_addr,
3355 ep->com.dev->rdev.lldi.rxq_ids[0]);
3357 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3361 err = net_xmit_errno(err);
3363 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3364 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3365 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3367 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3372 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3375 struct sockaddr_in *sin = (struct sockaddr_in *)
3376 &ep->com.local_addr;
3378 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3380 err = cxgb4_create_server_filter(
3381 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3382 sin->sin_addr.s_addr, sin->sin_port, 0,
3383 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3384 if (err == -EBUSY) {
3385 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3389 set_current_state(TASK_UNINTERRUPTIBLE);
3390 schedule_timeout(usecs_to_jiffies(100));
3392 } while (err == -EBUSY);
3394 c4iw_init_wr_wait(&ep->com.wr_wait);
3395 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3396 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3397 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3399 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3403 err = net_xmit_errno(err);
3406 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3408 &sin->sin_addr, ntohs(sin->sin_port));
3412 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3415 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3416 struct c4iw_listen_ep *ep;
3420 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3422 pr_err("%s - cannot alloc ep\n", __func__);
3426 skb_queue_head_init(&ep->com.ep_skb_list);
3427 pr_debug("%s ep %p\n", __func__, ep);
3428 ep->com.cm_id = cm_id;
3429 ref_cm_id(&ep->com);
3431 ep->backlog = backlog;
3432 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3433 sizeof(ep->com.local_addr));
3436 * Allocate a server TID.
3438 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3439 ep->com.local_addr.ss_family == AF_INET)
3440 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3441 cm_id->m_local_addr.ss_family, ep);
3443 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3444 cm_id->m_local_addr.ss_family, ep);
3446 if (ep->stid == -1) {
3447 pr_err("%s - cannot alloc stid\n", __func__);
3451 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3453 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3454 sizeof(ep->com.local_addr));
3456 state_set(&ep->com, LISTEN);
3457 if (ep->com.local_addr.ss_family == AF_INET)
3458 err = create_server4(dev, ep);
3460 err = create_server6(dev, ep);
3462 cm_id->provider_data = ep;
3466 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3467 ep->com.local_addr.ss_family);
3469 deref_cm_id(&ep->com);
3470 c4iw_put_ep(&ep->com);
3476 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3479 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3481 pr_debug("%s ep %p\n", __func__, ep);
3484 state_set(&ep->com, DEAD);
3485 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3486 ep->com.local_addr.ss_family == AF_INET) {
3487 err = cxgb4_remove_server_filter(
3488 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3489 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3491 struct sockaddr_in6 *sin6;
3492 c4iw_init_wr_wait(&ep->com.wr_wait);
3493 err = cxgb4_remove_server(
3494 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3495 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3498 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3500 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3501 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3502 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3504 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3505 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3506 ep->com.local_addr.ss_family);
3508 deref_cm_id(&ep->com);
3509 c4iw_put_ep(&ep->com);
3513 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3518 struct c4iw_rdev *rdev;
3520 mutex_lock(&ep->com.mutex);
3522 pr_debug("%s ep %p state %s, abrupt %d\n", __func__, ep,
3523 states[ep->com.state], abrupt);
3526 * Ref the ep here in case we have fatal errors causing the
3527 * ep to be released and freed.
3529 c4iw_get_ep(&ep->com);
3531 rdev = &ep->com.dev->rdev;
3532 if (c4iw_fatal_error(rdev)) {
3534 close_complete_upcall(ep, -EIO);
3535 ep->com.state = DEAD;
3537 switch (ep->com.state) {
3546 ep->com.state = ABORTING;
3548 ep->com.state = CLOSING;
3551 * if we close before we see the fw4_ack() then we fix
3552 * up the timer state since we're reusing it.
3555 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3556 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3561 set_bit(CLOSE_SENT, &ep->com.flags);
3564 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3567 (void)stop_ep_timer(ep);
3568 ep->com.state = ABORTING;
3570 ep->com.state = MORIBUND;
3576 pr_debug("%s ignoring disconnect ep %p state %u\n",
3577 __func__, ep, ep->com.state);
3586 set_bit(EP_DISC_ABORT, &ep->com.history);
3587 close_complete_upcall(ep, -ECONNRESET);
3588 ret = send_abort(ep);
3590 set_bit(EP_DISC_CLOSE, &ep->com.history);
3591 ret = send_halfclose(ep);
3594 set_bit(EP_DISC_FAIL, &ep->com.history);
3597 close_complete_upcall(ep, -EIO);
3600 struct c4iw_qp_attributes attrs;
3602 attrs.next_state = C4IW_QP_STATE_ERROR;
3603 ret = c4iw_modify_qp(ep->com.qp->rhp,
3605 C4IW_QP_ATTR_NEXT_STATE,
3608 pr_err("%s - qp <- error failed!\n",
3614 mutex_unlock(&ep->com.mutex);
3615 c4iw_put_ep(&ep->com);
3617 release_ep_resources(ep);
3621 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3622 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3625 int atid = be32_to_cpu(req->tid);
3627 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3628 (__force u32) req->tid);
3632 switch (req->retval) {
3634 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3635 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3636 send_fw_act_open_req(ep, atid);
3640 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3641 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3642 send_fw_act_open_req(ep, atid);
3647 pr_info("%s unexpected ofld conn wr retval %d\n",
3648 __func__, req->retval);
3651 pr_err("active ofld_connect_wr failure %d atid %d\n",
3653 mutex_lock(&dev->rdev.stats.lock);
3654 dev->rdev.stats.act_ofld_conn_fails++;
3655 mutex_unlock(&dev->rdev.stats.lock);
3656 connect_reply_upcall(ep, status2errno(req->retval));
3657 state_set(&ep->com, DEAD);
3658 if (ep->com.remote_addr.ss_family == AF_INET6) {
3659 struct sockaddr_in6 *sin6 =
3660 (struct sockaddr_in6 *)&ep->com.local_addr;
3661 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3662 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3664 remove_handle(dev, &dev->atid_idr, atid);
3665 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3666 dst_release(ep->dst);
3667 cxgb4_l2t_release(ep->l2t);
3668 c4iw_put_ep(&ep->com);
3671 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3672 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3674 struct sk_buff *rpl_skb;
3675 struct cpl_pass_accept_req *cpl;
3678 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3681 pr_debug("%s passive open failure %d\n", __func__, req->retval);
3682 mutex_lock(&dev->rdev.stats.lock);
3683 dev->rdev.stats.pas_ofld_conn_fails++;
3684 mutex_unlock(&dev->rdev.stats.lock);
3687 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3688 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3689 (__force u32) htonl(
3690 (__force u32) req->tid)));
3691 ret = pass_accept_req(dev, rpl_skb);
3698 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3700 struct cpl_fw6_msg *rpl = cplhdr(skb);
3701 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3703 switch (rpl->type) {
3705 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3707 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3708 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3709 switch (req->t_state) {
3711 active_ofld_conn_reply(dev, skb, req);
3714 passive_ofld_conn_reply(dev, skb, req);
3717 pr_err("%s unexpected ofld conn wr state %d\n",
3718 __func__, req->t_state);
3726 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3729 __be16 hdr_len, vlantag, len;
3731 int tcp_hdr_len, ip_hdr_len;
3733 struct cpl_rx_pkt *cpl = cplhdr(skb);
3734 struct cpl_pass_accept_req *req;
3735 struct tcp_options_received tmp_opt;
3736 struct c4iw_dev *dev;
3737 enum chip_type type;
3739 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3740 /* Store values from cpl_rx_pkt in temporary location. */
3741 vlantag = cpl->vlan;
3743 l2info = cpl->l2info;
3744 hdr_len = cpl->hdr_len;
3747 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3750 * We need to parse the TCP options from SYN packet.
3751 * to generate cpl_pass_accept_req.
3753 memset(&tmp_opt, 0, sizeof(tmp_opt));
3754 tcp_clear_options(&tmp_opt);
3755 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3757 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3758 memset(req, 0, sizeof(*req));
3759 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3760 SYN_MAC_IDX_V(RX_MACIDX_G(
3761 be32_to_cpu(l2info))) |
3763 type = dev->rdev.lldi.adapter_type;
3764 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3765 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3767 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3768 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3769 eth_hdr_len = is_t4(type) ?
3770 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3771 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3772 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3773 IP_HDR_LEN_V(ip_hdr_len) |
3774 ETH_HDR_LEN_V(eth_hdr_len));
3775 } else { /* T6 and later */
3776 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3777 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3778 T6_IP_HDR_LEN_V(ip_hdr_len) |
3779 T6_ETH_HDR_LEN_V(eth_hdr_len));
3781 req->vlan = vlantag;
3783 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3784 PASS_OPEN_TOS_V(tos));
3785 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3786 if (tmp_opt.wscale_ok)
3787 req->tcpopt.wsf = tmp_opt.snd_wscale;
3788 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3789 if (tmp_opt.sack_ok)
3790 req->tcpopt.sack = 1;
3791 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3795 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3796 __be32 laddr, __be16 lport,
3797 __be32 raddr, __be16 rport,
3798 u32 rcv_isn, u32 filter, u16 window,
3799 u32 rss_qid, u8 port_id)
3801 struct sk_buff *req_skb;
3802 struct fw_ofld_connection_wr *req;
3803 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3806 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3809 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3810 memset(req, 0, sizeof(*req));
3811 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3812 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3813 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3814 req->le.filter = (__force __be32) filter;
3815 req->le.lport = lport;
3816 req->le.pport = rport;
3817 req->le.u.ipv4.lip = laddr;
3818 req->le.u.ipv4.pip = raddr;
3819 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3820 req->tcb.rcv_adv = htons(window);
3821 req->tcb.t_state_to_astid =
3822 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3823 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3824 FW_OFLD_CONNECTION_WR_ASTID_V(
3825 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3828 * We store the qid in opt2 which will be used by the firmware
3829 * to send us the wr response.
3831 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3834 * We initialize the MSS index in TCB to 0xF.
3835 * So that when driver sends cpl_pass_accept_rpl
3836 * TCB picks up the correct value. If this was 0
3837 * TP will ignore any value > 0 for MSS index.
3839 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3840 req->cookie = (uintptr_t)skb;
3842 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3843 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3845 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3853 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3854 * messages when a filter is being used instead of server to
3855 * redirect a syn packet. When packets hit filter they are redirected
3856 * to the offload queue and driver tries to establish the connection
3857 * using firmware work request.
3859 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3862 unsigned int filter;
3863 struct ethhdr *eh = NULL;
3864 struct vlan_ethhdr *vlan_eh = NULL;
3866 struct tcphdr *tcph;
3867 struct rss_header *rss = (void *)skb->data;
3868 struct cpl_rx_pkt *cpl = (void *)skb->data;
3869 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3870 struct l2t_entry *e;
3871 struct dst_entry *dst;
3872 struct c4iw_ep *lep = NULL;
3874 struct port_info *pi;
3875 struct net_device *pdev;
3876 u16 rss_qid, eth_hdr_len;
3879 struct neighbour *neigh;
3881 /* Drop all non-SYN packets */
3882 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3886 * Drop all packets which did not hit the filter.
3887 * Unlikely to happen.
3889 if (!(rss->filter_hit && rss->filter_tid))
3893 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3895 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3897 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3899 pr_debug("%s connect request on invalid stid %d\n",
3904 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3906 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3909 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3912 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3915 pr_err("T%d Chip is not supported\n",
3916 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3920 if (eth_hdr_len == ETH_HLEN) {
3921 eh = (struct ethhdr *)(req + 1);
3922 iph = (struct iphdr *)(eh + 1);
3924 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3925 iph = (struct iphdr *)(vlan_eh + 1);
3926 skb->vlan_tci = ntohs(cpl->vlan);
3929 if (iph->version != 0x4)
3932 tcph = (struct tcphdr *)(iph + 1);
3933 skb_set_network_header(skb, (void *)iph - (void *)rss);
3934 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3937 pr_debug("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3938 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3939 ntohs(tcph->source), iph->tos);
3941 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3942 iph->daddr, iph->saddr, tcph->dest,
3943 tcph->source, iph->tos);
3945 pr_err("%s - failed to find dst entry!\n",
3949 neigh = dst_neigh_lookup_skb(dst, skb);
3952 pr_err("%s - failed to allocate neigh!\n",
3957 if (neigh->dev->flags & IFF_LOOPBACK) {
3958 pdev = ip_dev_find(&init_net, iph->daddr);
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);
3965 pdev = get_real_dev(neigh->dev);
3966 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3968 pi = (struct port_info *)netdev_priv(pdev);
3969 tx_chan = cxgb4_port_chan(pdev);
3971 neigh_release(neigh);
3973 pr_err("%s - failed to allocate l2t entry!\n",
3978 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3979 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3980 window = (__force u16) htons((__force u16)tcph->window);
3982 /* Calcuate filter portion for LE region. */
3983 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3984 dev->rdev.lldi.ports[0],
3988 * Synthesize the cpl_pass_accept_req. We have everything except the
3989 * TID. Once firmware sends a reply with TID we update the TID field
3990 * in cpl and pass it through the regular cpl_pass_accept_req path.
3992 build_cpl_pass_accept_req(skb, stid, iph->tos);
3993 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3994 tcph->source, ntohl(tcph->seq), filter, window,
3995 rss_qid, pi->port_id);
3996 cxgb4_l2t_release(e);
4001 c4iw_put_ep(&lep->com);
4006 * These are the real handlers that are called from a
4009 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4010 [CPL_ACT_ESTABLISH] = act_establish,
4011 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4012 [CPL_RX_DATA] = rx_data,
4013 [CPL_ABORT_RPL_RSS] = abort_rpl,
4014 [CPL_ABORT_RPL] = abort_rpl,
4015 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4016 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4017 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4018 [CPL_PASS_ESTABLISH] = pass_establish,
4019 [CPL_PEER_CLOSE] = peer_close,
4020 [CPL_ABORT_REQ_RSS] = peer_abort,
4021 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4022 [CPL_RDMA_TERMINATE] = terminate,
4023 [CPL_FW4_ACK] = fw4_ack,
4024 [CPL_FW6_MSG] = deferred_fw6_msg,
4025 [CPL_RX_PKT] = rx_pkt,
4026 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4027 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4030 static void process_timeout(struct c4iw_ep *ep)
4032 struct c4iw_qp_attributes attrs;
4035 mutex_lock(&ep->com.mutex);
4036 pr_debug("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4038 set_bit(TIMEDOUT, &ep->com.history);
4039 switch (ep->com.state) {
4041 connect_reply_upcall(ep, -ETIMEDOUT);
4050 if (ep->com.cm_id && ep->com.qp) {
4051 attrs.next_state = C4IW_QP_STATE_ERROR;
4052 c4iw_modify_qp(ep->com.qp->rhp,
4053 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4056 close_complete_upcall(ep, -ETIMEDOUT);
4062 * These states are expected if the ep timed out at the same
4063 * time as another thread was calling stop_ep_timer().
4064 * So we silently do nothing for these states.
4069 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4070 __func__, ep, ep->hwtid, ep->com.state);
4073 mutex_unlock(&ep->com.mutex);
4075 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4076 c4iw_put_ep(&ep->com);
4079 static void process_timedout_eps(void)
4083 spin_lock_irq(&timeout_lock);
4084 while (!list_empty(&timeout_list)) {
4085 struct list_head *tmp;
4087 tmp = timeout_list.next;
4091 spin_unlock_irq(&timeout_lock);
4092 ep = list_entry(tmp, struct c4iw_ep, entry);
4093 process_timeout(ep);
4094 spin_lock_irq(&timeout_lock);
4096 spin_unlock_irq(&timeout_lock);
4099 static void process_work(struct work_struct *work)
4101 struct sk_buff *skb = NULL;
4102 struct c4iw_dev *dev;
4103 struct cpl_act_establish *rpl;
4104 unsigned int opcode;
4107 process_timedout_eps();
4108 while ((skb = skb_dequeue(&rxq))) {
4110 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4111 opcode = rpl->ot.opcode;
4113 BUG_ON(!work_handlers[opcode]);
4114 ret = work_handlers[opcode](dev, skb);
4117 process_timedout_eps();
4121 static DECLARE_WORK(skb_work, process_work);
4123 static void ep_timeout(unsigned long arg)
4125 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4128 spin_lock(&timeout_lock);
4129 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4131 * Only insert if it is not already on the list.
4133 if (!ep->entry.next) {
4134 list_add_tail(&ep->entry, &timeout_list);
4138 spin_unlock(&timeout_lock);
4140 queue_work(workq, &skb_work);
4144 * All the CM events are handled on a work queue to have a safe context.
4146 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4150 * Save dev in the skb->cb area.
4152 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4155 * Queue the skb and schedule the worker thread.
4157 skb_queue_tail(&rxq, skb);
4158 queue_work(workq, &skb_work);
4162 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4164 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4166 if (rpl->status != CPL_ERR_NONE) {
4167 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4168 rpl->status, GET_TID(rpl));
4174 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4176 struct cpl_fw6_msg *rpl = cplhdr(skb);
4177 struct c4iw_wr_wait *wr_waitp;
4180 pr_debug("%s type %u\n", __func__, rpl->type);
4182 switch (rpl->type) {
4183 case FW6_TYPE_WR_RPL:
4184 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4185 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4186 pr_debug("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4188 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4192 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4196 pr_err("%s unexpected fw6 msg type %u\n",
4197 __func__, rpl->type);
4204 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4206 struct cpl_abort_req_rss *req = cplhdr(skb);
4208 unsigned int tid = GET_TID(req);
4210 ep = get_ep_from_tid(dev, tid);
4211 /* This EP will be dereferenced in peer_abort() */
4213 pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4217 if (cxgb_is_neg_adv(req->status)) {
4218 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n",
4219 __func__, ep->hwtid, req->status,
4220 neg_adv_str(req->status));
4223 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4226 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4233 * Most upcalls from the T4 Core go to sched() to
4234 * schedule the processing on a work queue.
4236 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4237 [CPL_ACT_ESTABLISH] = sched,
4238 [CPL_ACT_OPEN_RPL] = sched,
4239 [CPL_RX_DATA] = sched,
4240 [CPL_ABORT_RPL_RSS] = sched,
4241 [CPL_ABORT_RPL] = sched,
4242 [CPL_PASS_OPEN_RPL] = sched,
4243 [CPL_CLOSE_LISTSRV_RPL] = sched,
4244 [CPL_PASS_ACCEPT_REQ] = sched,
4245 [CPL_PASS_ESTABLISH] = sched,
4246 [CPL_PEER_CLOSE] = sched,
4247 [CPL_CLOSE_CON_RPL] = sched,
4248 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4249 [CPL_RDMA_TERMINATE] = sched,
4250 [CPL_FW4_ACK] = sched,
4251 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4252 [CPL_FW6_MSG] = fw6_msg,
4253 [CPL_RX_PKT] = sched
4256 int __init c4iw_cm_init(void)
4258 spin_lock_init(&timeout_lock);
4259 skb_queue_head_init(&rxq);
4261 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4268 void c4iw_cm_term(void)
4270 WARN_ON(!list_empty(&timeout_list));
4271 flush_workqueue(workq);
4272 destroy_workqueue(workq);