2 * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
50 #include <rdma/ib_addr.h>
52 #include <libcxgb_cm.h>
56 static char *states[] = {
73 module_param(nocong, int, 0644);
74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
76 static int enable_ecn;
77 module_param(enable_ecn, int, 0644);
78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
80 static int dack_mode = 1;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
84 uint c4iw_max_read_depth = 32;
85 module_param(c4iw_max_read_depth, int, 0644);
86 MODULE_PARM_DESC(c4iw_max_read_depth,
87 "Per-connection max ORD/IRD (default=32)");
89 static int enable_tcp_timestamps;
90 module_param(enable_tcp_timestamps, int, 0644);
91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
93 static int enable_tcp_sack;
94 module_param(enable_tcp_sack, int, 0644);
95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
97 static int enable_tcp_window_scaling = 1;
98 module_param(enable_tcp_window_scaling, int, 0644);
99 MODULE_PARM_DESC(enable_tcp_window_scaling,
100 "Enable tcp window scaling (default=1)");
103 module_param(c4iw_debug, int, 0644);
104 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
106 static int peer2peer = 1;
107 module_param(peer2peer, int, 0644);
108 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
110 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
111 module_param(p2p_type, int, 0644);
112 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
113 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
115 static int ep_timeout_secs = 60;
116 module_param(ep_timeout_secs, int, 0644);
117 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
118 "in seconds (default=60)");
120 static int mpa_rev = 2;
121 module_param(mpa_rev, int, 0644);
122 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
123 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
124 " compliant (default=2)");
126 static int markers_enabled;
127 module_param(markers_enabled, int, 0644);
128 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
130 static int crc_enabled = 1;
131 module_param(crc_enabled, int, 0644);
132 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
134 static int rcv_win = 256 * 1024;
135 module_param(rcv_win, int, 0644);
136 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
138 static int snd_win = 128 * 1024;
139 module_param(snd_win, int, 0644);
140 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
142 static struct workqueue_struct *workq;
144 static struct sk_buff_head rxq;
146 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
147 static void ep_timeout(unsigned long arg);
148 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
149 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
151 static LIST_HEAD(timeout_list);
152 static spinlock_t timeout_lock;
154 static void deref_cm_id(struct c4iw_ep_common *epc)
156 epc->cm_id->rem_ref(epc->cm_id);
158 set_bit(CM_ID_DEREFED, &epc->history);
161 static void ref_cm_id(struct c4iw_ep_common *epc)
163 set_bit(CM_ID_REFED, &epc->history);
164 epc->cm_id->add_ref(epc->cm_id);
167 static void deref_qp(struct c4iw_ep *ep)
169 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
170 clear_bit(QP_REFERENCED, &ep->com.flags);
171 set_bit(QP_DEREFED, &ep->com.history);
174 static void ref_qp(struct c4iw_ep *ep)
176 set_bit(QP_REFERENCED, &ep->com.flags);
177 set_bit(QP_REFED, &ep->com.history);
178 c4iw_qp_add_ref(&ep->com.qp->ibqp);
181 static void start_ep_timer(struct c4iw_ep *ep)
183 PDBG("%s ep %p\n", __func__, ep);
184 if (timer_pending(&ep->timer)) {
185 pr_err("%s timer already started! ep %p\n",
189 clear_bit(TIMEOUT, &ep->com.flags);
190 c4iw_get_ep(&ep->com);
191 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
192 ep->timer.data = (unsigned long)ep;
193 ep->timer.function = ep_timeout;
194 add_timer(&ep->timer);
197 static int stop_ep_timer(struct c4iw_ep *ep)
199 PDBG("%s ep %p stopping\n", __func__, ep);
200 del_timer_sync(&ep->timer);
201 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
202 c4iw_put_ep(&ep->com);
208 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
209 struct l2t_entry *l2e)
213 if (c4iw_fatal_error(rdev)) {
215 PDBG("%s - device in error state - dropping\n", __func__);
218 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
221 else if (error == NET_XMIT_DROP)
223 return error < 0 ? error : 0;
226 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
230 if (c4iw_fatal_error(rdev)) {
232 PDBG("%s - device in error state - dropping\n", __func__);
235 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
238 return error < 0 ? error : 0;
241 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
243 u32 len = roundup(sizeof(struct cpl_tid_release), 16);
245 skb = get_skb(skb, len, GFP_KERNEL);
249 cxgb_mk_tid_release(skb, len, hwtid, 0);
250 c4iw_ofld_send(rdev, skb);
254 static void set_emss(struct c4iw_ep *ep, u16 opt)
256 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
257 ((AF_INET == ep->com.remote_addr.ss_family) ?
258 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
259 sizeof(struct tcphdr);
261 if (TCPOPT_TSTAMP_G(opt))
262 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
266 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
267 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
268 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
272 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
274 enum c4iw_ep_state state;
276 mutex_lock(&epc->mutex);
278 mutex_unlock(&epc->mutex);
282 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
287 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
289 mutex_lock(&epc->mutex);
290 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
291 __state_set(epc, new);
292 mutex_unlock(&epc->mutex);
296 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
302 len = roundup(sizeof(union cpl_wr_size), 16);
303 for (i = 0; i < size; i++) {
304 skb = alloc_skb(len, GFP_KERNEL);
307 skb_queue_tail(ep_skb_list, skb);
311 skb_queue_purge(ep_skb_list);
315 static void *alloc_ep(int size, gfp_t gfp)
317 struct c4iw_ep_common *epc;
319 epc = kzalloc(size, gfp);
321 kref_init(&epc->kref);
322 mutex_init(&epc->mutex);
323 c4iw_init_wr_wait(&epc->wr_wait);
325 PDBG("%s alloc ep %p\n", __func__, epc);
329 static void remove_ep_tid(struct c4iw_ep *ep)
333 spin_lock_irqsave(&ep->com.dev->lock, flags);
334 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
335 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
338 static void insert_ep_tid(struct c4iw_ep *ep)
342 spin_lock_irqsave(&ep->com.dev->lock, flags);
343 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
344 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
348 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
350 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
355 spin_lock_irqsave(&dev->lock, flags);
356 ep = idr_find(&dev->hwtid_idr, tid);
358 c4iw_get_ep(&ep->com);
359 spin_unlock_irqrestore(&dev->lock, flags);
364 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
366 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
369 struct c4iw_listen_ep *ep;
372 spin_lock_irqsave(&dev->lock, flags);
373 ep = idr_find(&dev->stid_idr, stid);
375 c4iw_get_ep(&ep->com);
376 spin_unlock_irqrestore(&dev->lock, flags);
380 void _c4iw_free_ep(struct kref *kref)
384 ep = container_of(kref, struct c4iw_ep, com.kref);
385 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
386 if (test_bit(QP_REFERENCED, &ep->com.flags))
388 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
389 if (ep->com.remote_addr.ss_family == AF_INET6) {
390 struct sockaddr_in6 *sin6 =
391 (struct sockaddr_in6 *)
395 ep->com.dev->rdev.lldi.ports[0],
396 (const u32 *)&sin6->sin6_addr.s6_addr,
399 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
400 dst_release(ep->dst);
401 cxgb4_l2t_release(ep->l2t);
403 kfree_skb(ep->mpa_skb);
405 if (!skb_queue_empty(&ep->com.ep_skb_list))
406 skb_queue_purge(&ep->com.ep_skb_list);
410 static void release_ep_resources(struct c4iw_ep *ep)
412 set_bit(RELEASE_RESOURCES, &ep->com.flags);
415 * If we have a hwtid, then remove it from the idr table
416 * so lookups will no longer find this endpoint. Otherwise
417 * we have a race where one thread finds the ep ptr just
418 * before the other thread is freeing the ep memory.
422 c4iw_put_ep(&ep->com);
425 static int status2errno(int status)
430 case CPL_ERR_CONN_RESET:
432 case CPL_ERR_ARP_MISS:
433 return -EHOSTUNREACH;
434 case CPL_ERR_CONN_TIMEDOUT:
436 case CPL_ERR_TCAM_FULL:
438 case CPL_ERR_CONN_EXIST:
446 * Try and reuse skbs already allocated...
448 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
450 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
453 skb_reset_transport_header(skb);
455 skb = alloc_skb(len, gfp);
457 t4_set_arp_err_handler(skb, NULL, NULL);
461 static struct net_device *get_real_dev(struct net_device *egress_dev)
463 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
466 static void arp_failure_discard(void *handle, struct sk_buff *skb)
468 pr_err(MOD "ARP failure\n");
472 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
474 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
479 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
480 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
483 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
487 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
488 release_ep_resources(ep);
492 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
496 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
497 c4iw_put_ep(&ep->parent_ep->com);
498 release_ep_resources(ep);
503 * Fake up a special CPL opcode and call sched() so process_work() will call
504 * _put_ep_safe() in a safe context to free the ep resources. This is needed
505 * because ARP error handlers are called in an ATOMIC context, and
506 * _c4iw_free_ep() needs to block.
508 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
511 struct cpl_act_establish *rpl = cplhdr(skb);
513 /* Set our special ARP_FAILURE opcode */
514 rpl->ot.opcode = cpl;
517 * Save ep in the skb->cb area, after where sched() will save the dev
520 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
521 sched(ep->com.dev, skb);
524 /* Handle an ARP failure for an accept */
525 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
527 struct c4iw_ep *ep = handle;
529 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
532 __state_set(&ep->com, DEAD);
533 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
537 * Handle an ARP failure for an active open.
539 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
541 struct c4iw_ep *ep = handle;
543 printk(KERN_ERR MOD "ARP failure during connect\n");
544 connect_reply_upcall(ep, -EHOSTUNREACH);
545 __state_set(&ep->com, DEAD);
546 if (ep->com.remote_addr.ss_family == AF_INET6) {
547 struct sockaddr_in6 *sin6 =
548 (struct sockaddr_in6 *)&ep->com.local_addr;
549 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
550 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
552 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
553 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
554 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
558 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
561 static void abort_arp_failure(void *handle, struct sk_buff *skb)
564 struct c4iw_ep *ep = handle;
565 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
566 struct cpl_abort_req *req = cplhdr(skb);
568 PDBG("%s rdev %p\n", __func__, rdev);
569 req->cmd = CPL_ABORT_NO_RST;
570 ret = c4iw_ofld_send(rdev, skb);
572 __state_set(&ep->com, DEAD);
573 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
577 static int send_flowc(struct c4iw_ep *ep)
579 struct fw_flowc_wr *flowc;
580 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
582 u16 vlan = ep->l2t->vlan;
588 if (vlan == CPL_L2T_VLAN_NONE)
593 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
595 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
596 FW_FLOWC_WR_NPARAMS_V(nparams));
597 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
598 16)) | FW_WR_FLOWID_V(ep->hwtid));
600 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
601 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
602 (ep->com.dev->rdev.lldi.pf));
603 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
604 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
605 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
606 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
607 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
608 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
609 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
610 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
611 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
612 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
613 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
614 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
615 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
616 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
620 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
621 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
622 flowc->mnemval[8].val = cpu_to_be32(pri);
624 /* Pad WR to 16 byte boundary */
625 flowc->mnemval[8].mnemonic = 0;
626 flowc->mnemval[8].val = 0;
628 for (i = 0; i < 9; i++) {
629 flowc->mnemval[i].r4[0] = 0;
630 flowc->mnemval[i].r4[1] = 0;
631 flowc->mnemval[i].r4[2] = 0;
634 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
635 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
638 static int send_halfclose(struct c4iw_ep *ep)
640 struct cpl_close_con_req *req;
641 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
642 int wrlen = roundup(sizeof *req, 16);
644 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
648 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
649 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
650 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
651 memset(req, 0, wrlen);
652 INIT_TP_WR(req, ep->hwtid);
653 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
655 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
658 static int send_abort(struct c4iw_ep *ep)
660 struct cpl_abort_req *req;
661 int wrlen = roundup(sizeof *req, 16);
662 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
664 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
665 if (WARN_ON(!req_skb))
668 set_wr_txq(req_skb, CPL_PRIORITY_DATA, ep->txq_idx);
669 t4_set_arp_err_handler(req_skb, ep, abort_arp_failure);
670 req = (struct cpl_abort_req *)skb_put(req_skb, wrlen);
671 memset(req, 0, wrlen);
672 INIT_TP_WR(req, ep->hwtid);
673 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
674 req->cmd = CPL_ABORT_SEND_RST;
675 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
678 static int send_connect(struct c4iw_ep *ep)
680 struct cpl_act_open_req *req = NULL;
681 struct cpl_t5_act_open_req *t5req = NULL;
682 struct cpl_t6_act_open_req *t6req = NULL;
683 struct cpl_act_open_req6 *req6 = NULL;
684 struct cpl_t5_act_open_req6 *t5req6 = NULL;
685 struct cpl_t6_act_open_req6 *t6req6 = NULL;
689 unsigned int mtu_idx;
691 int win, sizev4, sizev6, wrlen;
692 struct sockaddr_in *la = (struct sockaddr_in *)
694 struct sockaddr_in *ra = (struct sockaddr_in *)
695 &ep->com.remote_addr;
696 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
698 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
699 &ep->com.remote_addr;
701 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
702 u32 isn = (prandom_u32() & ~7UL) - 1;
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 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
729 skb = get_skb(NULL, wrlen, GFP_KERNEL);
731 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
735 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
737 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
738 enable_tcp_timestamps,
739 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
740 wscale = cxgb_compute_wscale(rcv_win);
743 * Specify the largest window that will fit in opt0. The
744 * remainder will be specified in the rx_data_ack.
746 win = ep->rcv_win >> 10;
747 if (win > RCV_BUFSIZ_M)
750 opt0 = (nocong ? NO_CONG_F : 0) |
753 WND_SCALE_V(wscale) |
755 L2T_IDX_V(ep->l2t->idx) |
756 TX_CHAN_V(ep->tx_chan) |
757 SMAC_SEL_V(ep->smac_idx) |
758 DSCP_V(ep->tos >> 2) |
759 ULP_MODE_V(ULP_MODE_TCPDDP) |
761 opt2 = RX_CHANNEL_V(0) |
762 CCTRL_ECN_V(enable_ecn) |
763 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
764 if (enable_tcp_timestamps)
765 opt2 |= TSTAMPS_EN_F;
768 if (wscale && enable_tcp_window_scaling)
769 opt2 |= WND_SCALE_EN_F;
770 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
774 opt2 |= T5_OPT_2_VALID_F;
775 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
779 if (ep->com.remote_addr.ss_family == AF_INET6)
780 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
781 (const u32 *)&la6->sin6_addr.s6_addr, 1);
783 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
785 if (ep->com.remote_addr.ss_family == AF_INET) {
786 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
788 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
792 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
794 INIT_TP_WR(t5req, 0);
795 req = (struct cpl_act_open_req *)t5req;
798 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
800 INIT_TP_WR(t6req, 0);
801 req = (struct cpl_act_open_req *)t6req;
802 t5req = (struct cpl_t5_act_open_req *)t6req;
805 pr_err("T%d Chip is not supported\n",
806 CHELSIO_CHIP_VERSION(adapter_type));
811 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
812 ((ep->rss_qid<<14) | ep->atid)));
813 req->local_port = la->sin_port;
814 req->peer_port = ra->sin_port;
815 req->local_ip = la->sin_addr.s_addr;
816 req->peer_ip = ra->sin_addr.s_addr;
817 req->opt0 = cpu_to_be64(opt0);
819 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
820 req->params = cpu_to_be32(cxgb4_select_ntuple(
821 ep->com.dev->rdev.lldi.ports[0],
823 req->opt2 = cpu_to_be32(opt2);
825 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
827 ep->com.dev->rdev.lldi.ports[0],
829 t5req->rsvd = cpu_to_be32(isn);
830 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
831 t5req->opt2 = cpu_to_be32(opt2);
834 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
836 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
840 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
842 INIT_TP_WR(t5req6, 0);
843 req6 = (struct cpl_act_open_req6 *)t5req6;
846 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
848 INIT_TP_WR(t6req6, 0);
849 req6 = (struct cpl_act_open_req6 *)t6req6;
850 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
853 pr_err("T%d Chip is not supported\n",
854 CHELSIO_CHIP_VERSION(adapter_type));
859 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
860 ((ep->rss_qid<<14)|ep->atid)));
861 req6->local_port = la6->sin6_port;
862 req6->peer_port = ra6->sin6_port;
863 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
864 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
865 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
866 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
867 req6->opt0 = cpu_to_be64(opt0);
869 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
870 req6->params = cpu_to_be32(cxgb4_select_ntuple(
871 ep->com.dev->rdev.lldi.ports[0],
873 req6->opt2 = cpu_to_be32(opt2);
875 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
877 ep->com.dev->rdev.lldi.ports[0],
879 t5req6->rsvd = cpu_to_be32(isn);
880 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
881 t5req6->opt2 = cpu_to_be32(opt2);
885 set_bit(ACT_OPEN_REQ, &ep->com.history);
886 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
888 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
889 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
890 (const u32 *)&la6->sin6_addr.s6_addr, 1);
894 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
897 int mpalen, wrlen, ret;
898 struct fw_ofld_tx_data_wr *req;
899 struct mpa_message *mpa;
900 struct mpa_v2_conn_params mpa_v2_params;
902 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
904 BUG_ON(skb_cloned(skb));
906 mpalen = sizeof(*mpa) + ep->plen;
907 if (mpa_rev_to_use == 2)
908 mpalen += sizeof(struct mpa_v2_conn_params);
909 wrlen = roundup(mpalen + sizeof *req, 16);
910 skb = get_skb(skb, wrlen, GFP_KERNEL);
912 connect_reply_upcall(ep, -ENOMEM);
915 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
917 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
918 memset(req, 0, wrlen);
919 req->op_to_immdlen = cpu_to_be32(
920 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
922 FW_WR_IMMDLEN_V(mpalen));
923 req->flowid_len16 = cpu_to_be32(
924 FW_WR_FLOWID_V(ep->hwtid) |
925 FW_WR_LEN16_V(wrlen >> 4));
926 req->plen = cpu_to_be32(mpalen);
927 req->tunnel_to_proxy = cpu_to_be32(
928 FW_OFLD_TX_DATA_WR_FLUSH_F |
929 FW_OFLD_TX_DATA_WR_SHOVE_F);
931 mpa = (struct mpa_message *)(req + 1);
932 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
936 mpa->flags |= MPA_CRC;
937 if (markers_enabled) {
938 mpa->flags |= MPA_MARKERS;
939 ep->mpa_attr.recv_marker_enabled = 1;
941 ep->mpa_attr.recv_marker_enabled = 0;
943 if (mpa_rev_to_use == 2)
944 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
946 mpa->private_data_size = htons(ep->plen);
947 mpa->revision = mpa_rev_to_use;
948 if (mpa_rev_to_use == 1) {
949 ep->tried_with_mpa_v1 = 1;
950 ep->retry_with_mpa_v1 = 0;
953 if (mpa_rev_to_use == 2) {
954 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
955 sizeof (struct mpa_v2_conn_params));
956 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
958 mpa_v2_params.ird = htons((u16)ep->ird);
959 mpa_v2_params.ord = htons((u16)ep->ord);
962 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
963 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
965 htons(MPA_V2_RDMA_WRITE_RTR);
966 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
968 htons(MPA_V2_RDMA_READ_RTR);
970 memcpy(mpa->private_data, &mpa_v2_params,
971 sizeof(struct mpa_v2_conn_params));
974 memcpy(mpa->private_data +
975 sizeof(struct mpa_v2_conn_params),
976 ep->mpa_pkt + sizeof(*mpa), ep->plen);
979 memcpy(mpa->private_data,
980 ep->mpa_pkt + sizeof(*mpa), ep->plen);
983 * Reference the mpa skb. This ensures the data area
984 * will remain in memory until the hw acks the tx.
985 * Function fw4_ack() will deref it.
988 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
991 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
995 __state_set(&ep->com, MPA_REQ_SENT);
996 ep->mpa_attr.initiator = 1;
997 ep->snd_seq += mpalen;
1001 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1004 struct fw_ofld_tx_data_wr *req;
1005 struct mpa_message *mpa;
1006 struct sk_buff *skb;
1007 struct mpa_v2_conn_params mpa_v2_params;
1009 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1011 mpalen = sizeof(*mpa) + plen;
1012 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1013 mpalen += sizeof(struct mpa_v2_conn_params);
1014 wrlen = roundup(mpalen + sizeof *req, 16);
1016 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1018 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1021 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1023 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1024 memset(req, 0, wrlen);
1025 req->op_to_immdlen = cpu_to_be32(
1026 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1028 FW_WR_IMMDLEN_V(mpalen));
1029 req->flowid_len16 = cpu_to_be32(
1030 FW_WR_FLOWID_V(ep->hwtid) |
1031 FW_WR_LEN16_V(wrlen >> 4));
1032 req->plen = cpu_to_be32(mpalen);
1033 req->tunnel_to_proxy = cpu_to_be32(
1034 FW_OFLD_TX_DATA_WR_FLUSH_F |
1035 FW_OFLD_TX_DATA_WR_SHOVE_F);
1037 mpa = (struct mpa_message *)(req + 1);
1038 memset(mpa, 0, sizeof(*mpa));
1039 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1040 mpa->flags = MPA_REJECT;
1041 mpa->revision = ep->mpa_attr.version;
1042 mpa->private_data_size = htons(plen);
1044 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1045 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1046 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1047 sizeof (struct mpa_v2_conn_params));
1048 mpa_v2_params.ird = htons(((u16)ep->ird) |
1049 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1051 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1053 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1054 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1055 FW_RI_INIT_P2PTYPE_READ_REQ ?
1056 MPA_V2_RDMA_READ_RTR : 0) : 0));
1057 memcpy(mpa->private_data, &mpa_v2_params,
1058 sizeof(struct mpa_v2_conn_params));
1061 memcpy(mpa->private_data +
1062 sizeof(struct mpa_v2_conn_params), pdata, plen);
1065 memcpy(mpa->private_data, pdata, plen);
1068 * Reference the mpa skb again. This ensures the data area
1069 * will remain in memory until the hw acks the tx.
1070 * Function fw4_ack() will deref it.
1073 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1074 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1075 BUG_ON(ep->mpa_skb);
1077 ep->snd_seq += mpalen;
1078 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1081 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1084 struct fw_ofld_tx_data_wr *req;
1085 struct mpa_message *mpa;
1086 struct sk_buff *skb;
1087 struct mpa_v2_conn_params mpa_v2_params;
1089 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1091 mpalen = sizeof(*mpa) + plen;
1092 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1093 mpalen += sizeof(struct mpa_v2_conn_params);
1094 wrlen = roundup(mpalen + sizeof *req, 16);
1096 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1098 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1101 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1103 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1104 memset(req, 0, wrlen);
1105 req->op_to_immdlen = cpu_to_be32(
1106 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1108 FW_WR_IMMDLEN_V(mpalen));
1109 req->flowid_len16 = cpu_to_be32(
1110 FW_WR_FLOWID_V(ep->hwtid) |
1111 FW_WR_LEN16_V(wrlen >> 4));
1112 req->plen = cpu_to_be32(mpalen);
1113 req->tunnel_to_proxy = cpu_to_be32(
1114 FW_OFLD_TX_DATA_WR_FLUSH_F |
1115 FW_OFLD_TX_DATA_WR_SHOVE_F);
1117 mpa = (struct mpa_message *)(req + 1);
1118 memset(mpa, 0, sizeof(*mpa));
1119 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1121 if (ep->mpa_attr.crc_enabled)
1122 mpa->flags |= MPA_CRC;
1123 if (ep->mpa_attr.recv_marker_enabled)
1124 mpa->flags |= MPA_MARKERS;
1125 mpa->revision = ep->mpa_attr.version;
1126 mpa->private_data_size = htons(plen);
1128 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1129 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1130 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1131 sizeof (struct mpa_v2_conn_params));
1132 mpa_v2_params.ird = htons((u16)ep->ird);
1133 mpa_v2_params.ord = htons((u16)ep->ord);
1134 if (peer2peer && (ep->mpa_attr.p2p_type !=
1135 FW_RI_INIT_P2PTYPE_DISABLED)) {
1136 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1138 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1139 mpa_v2_params.ord |=
1140 htons(MPA_V2_RDMA_WRITE_RTR);
1141 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1142 mpa_v2_params.ord |=
1143 htons(MPA_V2_RDMA_READ_RTR);
1146 memcpy(mpa->private_data, &mpa_v2_params,
1147 sizeof(struct mpa_v2_conn_params));
1150 memcpy(mpa->private_data +
1151 sizeof(struct mpa_v2_conn_params), pdata, plen);
1154 memcpy(mpa->private_data, pdata, plen);
1157 * Reference the mpa skb. This ensures the data area
1158 * will remain in memory until the hw acks the tx.
1159 * Function fw4_ack() will deref it.
1162 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1164 __state_set(&ep->com, MPA_REP_SENT);
1165 ep->snd_seq += mpalen;
1166 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1169 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1172 struct cpl_act_establish *req = cplhdr(skb);
1173 unsigned int tid = GET_TID(req);
1174 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1175 struct tid_info *t = dev->rdev.lldi.tids;
1178 ep = lookup_atid(t, atid);
1180 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1181 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1183 mutex_lock(&ep->com.mutex);
1184 dst_confirm(ep->dst);
1186 /* setup the hwtid for this connection */
1188 cxgb4_insert_tid(t, ep, tid);
1191 ep->snd_seq = be32_to_cpu(req->snd_isn);
1192 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1194 set_emss(ep, ntohs(req->tcp_opt));
1196 /* dealloc the atid */
1197 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1198 cxgb4_free_atid(t, atid);
1199 set_bit(ACT_ESTAB, &ep->com.history);
1201 /* start MPA negotiation */
1202 ret = send_flowc(ep);
1205 if (ep->retry_with_mpa_v1)
1206 ret = send_mpa_req(ep, skb, 1);
1208 ret = send_mpa_req(ep, skb, mpa_rev);
1211 mutex_unlock(&ep->com.mutex);
1214 mutex_unlock(&ep->com.mutex);
1215 connect_reply_upcall(ep, -ENOMEM);
1216 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1220 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1222 struct iw_cm_event event;
1224 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1225 memset(&event, 0, sizeof(event));
1226 event.event = IW_CM_EVENT_CLOSE;
1227 event.status = status;
1228 if (ep->com.cm_id) {
1229 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1230 ep, ep->com.cm_id, ep->hwtid);
1231 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1232 deref_cm_id(&ep->com);
1233 set_bit(CLOSE_UPCALL, &ep->com.history);
1237 static void peer_close_upcall(struct c4iw_ep *ep)
1239 struct iw_cm_event event;
1241 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1242 memset(&event, 0, sizeof(event));
1243 event.event = IW_CM_EVENT_DISCONNECT;
1244 if (ep->com.cm_id) {
1245 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1246 ep, ep->com.cm_id, ep->hwtid);
1247 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1248 set_bit(DISCONN_UPCALL, &ep->com.history);
1252 static void peer_abort_upcall(struct c4iw_ep *ep)
1254 struct iw_cm_event event;
1256 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1257 memset(&event, 0, sizeof(event));
1258 event.event = IW_CM_EVENT_CLOSE;
1259 event.status = -ECONNRESET;
1260 if (ep->com.cm_id) {
1261 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1262 ep->com.cm_id, ep->hwtid);
1263 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1264 deref_cm_id(&ep->com);
1265 set_bit(ABORT_UPCALL, &ep->com.history);
1269 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1271 struct iw_cm_event event;
1273 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1274 memset(&event, 0, sizeof(event));
1275 event.event = IW_CM_EVENT_CONNECT_REPLY;
1276 event.status = status;
1277 memcpy(&event.local_addr, &ep->com.local_addr,
1278 sizeof(ep->com.local_addr));
1279 memcpy(&event.remote_addr, &ep->com.remote_addr,
1280 sizeof(ep->com.remote_addr));
1282 if ((status == 0) || (status == -ECONNREFUSED)) {
1283 if (!ep->tried_with_mpa_v1) {
1284 /* this means MPA_v2 is used */
1285 event.ord = ep->ird;
1286 event.ird = ep->ord;
1287 event.private_data_len = ep->plen -
1288 sizeof(struct mpa_v2_conn_params);
1289 event.private_data = ep->mpa_pkt +
1290 sizeof(struct mpa_message) +
1291 sizeof(struct mpa_v2_conn_params);
1293 /* this means MPA_v1 is used */
1294 event.ord = cur_max_read_depth(ep->com.dev);
1295 event.ird = cur_max_read_depth(ep->com.dev);
1296 event.private_data_len = ep->plen;
1297 event.private_data = ep->mpa_pkt +
1298 sizeof(struct mpa_message);
1302 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1304 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1305 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1308 deref_cm_id(&ep->com);
1311 static int connect_request_upcall(struct c4iw_ep *ep)
1313 struct iw_cm_event event;
1316 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1317 memset(&event, 0, sizeof(event));
1318 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1319 memcpy(&event.local_addr, &ep->com.local_addr,
1320 sizeof(ep->com.local_addr));
1321 memcpy(&event.remote_addr, &ep->com.remote_addr,
1322 sizeof(ep->com.remote_addr));
1323 event.provider_data = ep;
1324 if (!ep->tried_with_mpa_v1) {
1325 /* this means MPA_v2 is used */
1326 event.ord = ep->ord;
1327 event.ird = ep->ird;
1328 event.private_data_len = ep->plen -
1329 sizeof(struct mpa_v2_conn_params);
1330 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1331 sizeof(struct mpa_v2_conn_params);
1333 /* this means MPA_v1 is used. Send max supported */
1334 event.ord = cur_max_read_depth(ep->com.dev);
1335 event.ird = cur_max_read_depth(ep->com.dev);
1336 event.private_data_len = ep->plen;
1337 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1339 c4iw_get_ep(&ep->com);
1340 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1343 c4iw_put_ep(&ep->com);
1344 set_bit(CONNREQ_UPCALL, &ep->com.history);
1345 c4iw_put_ep(&ep->parent_ep->com);
1349 static void established_upcall(struct c4iw_ep *ep)
1351 struct iw_cm_event event;
1353 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1354 memset(&event, 0, sizeof(event));
1355 event.event = IW_CM_EVENT_ESTABLISHED;
1356 event.ird = ep->ord;
1357 event.ord = ep->ird;
1358 if (ep->com.cm_id) {
1359 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1360 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1361 set_bit(ESTAB_UPCALL, &ep->com.history);
1365 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1367 struct cpl_rx_data_ack *req;
1368 struct sk_buff *skb;
1369 int wrlen = roundup(sizeof *req, 16);
1371 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1372 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1374 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1379 * If we couldn't specify the entire rcv window at connection setup
1380 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1381 * then add the overage in to the credits returned.
1383 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1384 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1386 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1387 memset(req, 0, wrlen);
1388 INIT_TP_WR(req, ep->hwtid);
1389 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1391 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
1393 RX_DACK_MODE_V(dack_mode));
1394 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1395 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1399 #define RELAXED_IRD_NEGOTIATION 1
1402 * process_mpa_reply - process streaming mode MPA reply
1406 * 0 upon success indicating a connect request was delivered to the ULP
1407 * or the mpa request is incomplete but valid so far.
1409 * 1 if a failure requires the caller to close the connection.
1411 * 2 if a failure requires the caller to abort the connection.
1413 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1415 struct mpa_message *mpa;
1416 struct mpa_v2_conn_params *mpa_v2_params;
1418 u16 resp_ird, resp_ord;
1419 u8 rtr_mismatch = 0, insuff_ird = 0;
1420 struct c4iw_qp_attributes attrs;
1421 enum c4iw_qp_attr_mask mask;
1425 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1428 * If we get more than the supported amount of private data
1429 * then we must fail this connection.
1431 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1433 goto err_stop_timer;
1437 * copy the new data into our accumulation buffer.
1439 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1441 ep->mpa_pkt_len += skb->len;
1444 * if we don't even have the mpa message, then bail.
1446 if (ep->mpa_pkt_len < sizeof(*mpa))
1448 mpa = (struct mpa_message *) ep->mpa_pkt;
1450 /* Validate MPA header. */
1451 if (mpa->revision > mpa_rev) {
1452 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1453 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1455 goto err_stop_timer;
1457 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1459 goto err_stop_timer;
1462 plen = ntohs(mpa->private_data_size);
1465 * Fail if there's too much private data.
1467 if (plen > MPA_MAX_PRIVATE_DATA) {
1469 goto err_stop_timer;
1473 * If plen does not account for pkt size
1475 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1477 goto err_stop_timer;
1480 ep->plen = (u8) plen;
1483 * If we don't have all the pdata yet, then bail.
1484 * We'll continue process when more data arrives.
1486 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1489 if (mpa->flags & MPA_REJECT) {
1490 err = -ECONNREFUSED;
1491 goto err_stop_timer;
1495 * Stop mpa timer. If it expired, then
1496 * we ignore the MPA reply. process_timeout()
1497 * will abort the connection.
1499 if (stop_ep_timer(ep))
1503 * If we get here we have accumulated the entire mpa
1504 * start reply message including private data. And
1505 * the MPA header is valid.
1507 __state_set(&ep->com, FPDU_MODE);
1508 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1509 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1510 ep->mpa_attr.version = mpa->revision;
1511 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1513 if (mpa->revision == 2) {
1514 ep->mpa_attr.enhanced_rdma_conn =
1515 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1516 if (ep->mpa_attr.enhanced_rdma_conn) {
1517 mpa_v2_params = (struct mpa_v2_conn_params *)
1518 (ep->mpa_pkt + sizeof(*mpa));
1519 resp_ird = ntohs(mpa_v2_params->ird) &
1520 MPA_V2_IRD_ORD_MASK;
1521 resp_ord = ntohs(mpa_v2_params->ord) &
1522 MPA_V2_IRD_ORD_MASK;
1523 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1524 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1527 * This is a double-check. Ideally, below checks are
1528 * not required since ird/ord stuff has been taken
1529 * care of in c4iw_accept_cr
1531 if (ep->ird < resp_ord) {
1532 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1533 ep->com.dev->rdev.lldi.max_ordird_qp)
1537 } else if (ep->ird > resp_ord) {
1540 if (ep->ord > resp_ird) {
1541 if (RELAXED_IRD_NEGOTIATION)
1552 if (ntohs(mpa_v2_params->ird) &
1553 MPA_V2_PEER2PEER_MODEL) {
1554 if (ntohs(mpa_v2_params->ord) &
1555 MPA_V2_RDMA_WRITE_RTR)
1556 ep->mpa_attr.p2p_type =
1557 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1558 else if (ntohs(mpa_v2_params->ord) &
1559 MPA_V2_RDMA_READ_RTR)
1560 ep->mpa_attr.p2p_type =
1561 FW_RI_INIT_P2PTYPE_READ_REQ;
1564 } else if (mpa->revision == 1)
1566 ep->mpa_attr.p2p_type = p2p_type;
1568 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1569 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1570 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1571 ep->mpa_attr.recv_marker_enabled,
1572 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1573 ep->mpa_attr.p2p_type, p2p_type);
1576 * If responder's RTR does not match with that of initiator, assign
1577 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1578 * generated when moving QP to RTS state.
1579 * A TERM message will be sent after QP has moved to RTS state
1581 if ((ep->mpa_attr.version == 2) && peer2peer &&
1582 (ep->mpa_attr.p2p_type != p2p_type)) {
1583 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1587 attrs.mpa_attr = ep->mpa_attr;
1588 attrs.max_ird = ep->ird;
1589 attrs.max_ord = ep->ord;
1590 attrs.llp_stream_handle = ep;
1591 attrs.next_state = C4IW_QP_STATE_RTS;
1593 mask = C4IW_QP_ATTR_NEXT_STATE |
1594 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1595 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1597 /* bind QP and TID with INIT_WR */
1598 err = c4iw_modify_qp(ep->com.qp->rhp,
1599 ep->com.qp, mask, &attrs, 1);
1604 * If responder's RTR requirement did not match with what initiator
1605 * supports, generate TERM message
1608 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1609 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1610 attrs.ecode = MPA_NOMATCH_RTR;
1611 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1612 attrs.send_term = 1;
1613 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1614 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1621 * Generate TERM if initiator IRD is not sufficient for responder
1622 * provided ORD. Currently, we do the same behaviour even when
1623 * responder provided IRD is also not sufficient as regards to
1627 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1629 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1630 attrs.ecode = MPA_INSUFF_IRD;
1631 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1632 attrs.send_term = 1;
1633 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1634 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1645 connect_reply_upcall(ep, err);
1650 * process_mpa_request - process streaming mode MPA request
1654 * 0 upon success indicating a connect request was delivered to the ULP
1655 * or the mpa request is incomplete but valid so far.
1657 * 1 if a failure requires the caller to close the connection.
1659 * 2 if a failure requires the caller to abort the connection.
1661 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1663 struct mpa_message *mpa;
1664 struct mpa_v2_conn_params *mpa_v2_params;
1667 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1670 * If we get more than the supported amount of private data
1671 * then we must fail this connection.
1673 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1674 goto err_stop_timer;
1676 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1679 * Copy the new data into our accumulation buffer.
1681 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1683 ep->mpa_pkt_len += skb->len;
1686 * If we don't even have the mpa message, then bail.
1687 * We'll continue process when more data arrives.
1689 if (ep->mpa_pkt_len < sizeof(*mpa))
1692 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1693 mpa = (struct mpa_message *) ep->mpa_pkt;
1696 * Validate MPA Header.
1698 if (mpa->revision > mpa_rev) {
1699 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1700 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1701 goto err_stop_timer;
1704 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1705 goto err_stop_timer;
1707 plen = ntohs(mpa->private_data_size);
1710 * Fail if there's too much private data.
1712 if (plen > MPA_MAX_PRIVATE_DATA)
1713 goto err_stop_timer;
1716 * If plen does not account for pkt size
1718 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1719 goto err_stop_timer;
1720 ep->plen = (u8) plen;
1723 * If we don't have all the pdata yet, then bail.
1725 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1729 * If we get here we have accumulated the entire mpa
1730 * start reply message including private data.
1732 ep->mpa_attr.initiator = 0;
1733 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1734 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1735 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1736 ep->mpa_attr.version = mpa->revision;
1737 if (mpa->revision == 1)
1738 ep->tried_with_mpa_v1 = 1;
1739 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1741 if (mpa->revision == 2) {
1742 ep->mpa_attr.enhanced_rdma_conn =
1743 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1744 if (ep->mpa_attr.enhanced_rdma_conn) {
1745 mpa_v2_params = (struct mpa_v2_conn_params *)
1746 (ep->mpa_pkt + sizeof(*mpa));
1747 ep->ird = ntohs(mpa_v2_params->ird) &
1748 MPA_V2_IRD_ORD_MASK;
1749 ep->ird = min_t(u32, ep->ird,
1750 cur_max_read_depth(ep->com.dev));
1751 ep->ord = ntohs(mpa_v2_params->ord) &
1752 MPA_V2_IRD_ORD_MASK;
1753 ep->ord = min_t(u32, ep->ord,
1754 cur_max_read_depth(ep->com.dev));
1755 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1757 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1759 if (ntohs(mpa_v2_params->ord) &
1760 MPA_V2_RDMA_WRITE_RTR)
1761 ep->mpa_attr.p2p_type =
1762 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1763 else if (ntohs(mpa_v2_params->ord) &
1764 MPA_V2_RDMA_READ_RTR)
1765 ep->mpa_attr.p2p_type =
1766 FW_RI_INIT_P2PTYPE_READ_REQ;
1769 } else if (mpa->revision == 1)
1771 ep->mpa_attr.p2p_type = p2p_type;
1773 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1774 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1775 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1776 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1777 ep->mpa_attr.p2p_type);
1779 __state_set(&ep->com, MPA_REQ_RCVD);
1782 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1783 if (ep->parent_ep->com.state != DEAD) {
1784 if (connect_request_upcall(ep))
1785 goto err_unlock_parent;
1787 goto err_unlock_parent;
1789 mutex_unlock(&ep->parent_ep->com.mutex);
1793 mutex_unlock(&ep->parent_ep->com.mutex);
1796 (void)stop_ep_timer(ep);
1801 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1804 struct cpl_rx_data *hdr = cplhdr(skb);
1805 unsigned int dlen = ntohs(hdr->len);
1806 unsigned int tid = GET_TID(hdr);
1807 __u8 status = hdr->status;
1810 ep = get_ep_from_tid(dev, tid);
1813 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1814 skb_pull(skb, sizeof(*hdr));
1815 skb_trim(skb, dlen);
1816 mutex_lock(&ep->com.mutex);
1818 /* update RX credits */
1819 update_rx_credits(ep, dlen);
1821 switch (ep->com.state) {
1823 ep->rcv_seq += dlen;
1824 disconnect = process_mpa_reply(ep, skb);
1827 ep->rcv_seq += dlen;
1828 disconnect = process_mpa_request(ep, skb);
1831 struct c4iw_qp_attributes attrs;
1832 BUG_ON(!ep->com.qp);
1834 pr_err("%s Unexpected streaming data." \
1835 " qpid %u ep %p state %d tid %u status %d\n",
1836 __func__, ep->com.qp->wq.sq.qid, ep,
1837 ep->com.state, ep->hwtid, status);
1838 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1839 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1840 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1847 mutex_unlock(&ep->com.mutex);
1849 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1850 c4iw_put_ep(&ep->com);
1854 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1857 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1859 unsigned int tid = GET_TID(rpl);
1861 ep = get_ep_from_tid(dev, tid);
1863 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1866 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1867 mutex_lock(&ep->com.mutex);
1868 switch (ep->com.state) {
1870 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1871 __state_set(&ep->com, DEAD);
1875 printk(KERN_ERR "%s ep %p state %d\n",
1876 __func__, ep, ep->com.state);
1879 mutex_unlock(&ep->com.mutex);
1882 release_ep_resources(ep);
1883 c4iw_put_ep(&ep->com);
1887 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1889 struct sk_buff *skb;
1890 struct fw_ofld_connection_wr *req;
1891 unsigned int mtu_idx;
1893 struct sockaddr_in *sin;
1896 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1897 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1898 memset(req, 0, sizeof(*req));
1899 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1900 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1901 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1902 ep->com.dev->rdev.lldi.ports[0],
1904 sin = (struct sockaddr_in *)&ep->com.local_addr;
1905 req->le.lport = sin->sin_port;
1906 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1907 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1908 req->le.pport = sin->sin_port;
1909 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1910 req->tcb.t_state_to_astid =
1911 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1912 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1913 req->tcb.cplrxdataack_cplpassacceptrpl =
1914 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1915 req->tcb.tx_max = (__force __be32) jiffies;
1916 req->tcb.rcv_adv = htons(1);
1917 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1918 enable_tcp_timestamps,
1919 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1920 wscale = cxgb_compute_wscale(rcv_win);
1923 * Specify the largest window that will fit in opt0. The
1924 * remainder will be specified in the rx_data_ack.
1926 win = ep->rcv_win >> 10;
1927 if (win > RCV_BUFSIZ_M)
1930 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1931 (nocong ? NO_CONG_F : 0) |
1934 WND_SCALE_V(wscale) |
1935 MSS_IDX_V(mtu_idx) |
1936 L2T_IDX_V(ep->l2t->idx) |
1937 TX_CHAN_V(ep->tx_chan) |
1938 SMAC_SEL_V(ep->smac_idx) |
1939 DSCP_V(ep->tos >> 2) |
1940 ULP_MODE_V(ULP_MODE_TCPDDP) |
1942 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1943 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1945 CCTRL_ECN_V(enable_ecn) |
1946 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1947 if (enable_tcp_timestamps)
1948 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1949 if (enable_tcp_sack)
1950 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1951 if (wscale && enable_tcp_window_scaling)
1952 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1953 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1954 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1955 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1956 set_bit(ACT_OFLD_CONN, &ep->com.history);
1957 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1961 * Some of the error codes above implicitly indicate that there is no TID
1962 * allocated with the result of an ACT_OPEN. We use this predicate to make
1965 static inline int act_open_has_tid(int status)
1967 return (status != CPL_ERR_TCAM_PARITY &&
1968 status != CPL_ERR_TCAM_MISS &&
1969 status != CPL_ERR_TCAM_FULL &&
1970 status != CPL_ERR_CONN_EXIST_SYNRECV &&
1971 status != CPL_ERR_CONN_EXIST);
1974 static char *neg_adv_str(unsigned int status)
1977 case CPL_ERR_RTX_NEG_ADVICE:
1978 return "Retransmit timeout";
1979 case CPL_ERR_PERSIST_NEG_ADVICE:
1980 return "Persist timeout";
1981 case CPL_ERR_KEEPALV_NEG_ADVICE:
1982 return "Keepalive timeout";
1988 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1990 ep->snd_win = snd_win;
1991 ep->rcv_win = rcv_win;
1992 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
1995 #define ACT_OPEN_RETRY_COUNT 2
1997 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1998 struct dst_entry *dst, struct c4iw_dev *cdev,
1999 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2001 struct neighbour *n;
2003 struct net_device *pdev;
2005 n = dst_neigh_lookup(dst, peer_ip);
2011 if (n->dev->flags & IFF_LOOPBACK) {
2013 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2014 else if (IS_ENABLED(CONFIG_IPV6))
2015 for_each_netdev(&init_net, pdev) {
2016 if (ipv6_chk_addr(&init_net,
2017 (struct in6_addr *)peer_ip,
2028 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2029 n, pdev, rt_tos2priority(tos));
2032 ep->mtu = pdev->mtu;
2033 ep->tx_chan = cxgb4_port_chan(pdev);
2034 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2035 cxgb4_port_viid(pdev));
2036 step = cdev->rdev.lldi.ntxq /
2037 cdev->rdev.lldi.nchan;
2038 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2039 step = cdev->rdev.lldi.nrxq /
2040 cdev->rdev.lldi.nchan;
2041 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2042 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2043 cxgb4_port_idx(pdev) * step];
2044 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2047 pdev = get_real_dev(n->dev);
2048 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2052 ep->mtu = dst_mtu(dst);
2053 ep->tx_chan = cxgb4_port_chan(pdev);
2054 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2055 cxgb4_port_viid(pdev));
2056 step = cdev->rdev.lldi.ntxq /
2057 cdev->rdev.lldi.nchan;
2058 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2059 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2060 step = cdev->rdev.lldi.nrxq /
2061 cdev->rdev.lldi.nchan;
2062 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2063 cxgb4_port_idx(pdev) * step];
2064 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2067 ep->retry_with_mpa_v1 = 0;
2068 ep->tried_with_mpa_v1 = 0;
2080 static int c4iw_reconnect(struct c4iw_ep *ep)
2084 struct sockaddr_in *laddr = (struct sockaddr_in *)
2085 &ep->com.cm_id->m_local_addr;
2086 struct sockaddr_in *raddr = (struct sockaddr_in *)
2087 &ep->com.cm_id->m_remote_addr;
2088 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2089 &ep->com.cm_id->m_local_addr;
2090 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2091 &ep->com.cm_id->m_remote_addr;
2095 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2096 init_timer(&ep->timer);
2097 c4iw_init_wr_wait(&ep->com.wr_wait);
2099 /* When MPA revision is different on nodes, the node with MPA_rev=2
2100 * tries to reconnect with MPA_rev 1 for the same EP through
2101 * c4iw_reconnect(), where the same EP is assigned with new tid for
2102 * further connection establishment. As we are using the same EP pointer
2103 * for reconnect, few skbs are used during the previous c4iw_connect(),
2104 * which leaves the EP with inadequate skbs for further
2105 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2106 * skb_list() during peer_abort(). Allocate skbs which is already used.
2108 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2109 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2115 * Allocate an active TID to initiate a TCP connection.
2117 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2118 if (ep->atid == -1) {
2119 pr_err("%s - cannot alloc atid.\n", __func__);
2123 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2126 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2127 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2128 laddr->sin_addr.s_addr,
2129 raddr->sin_addr.s_addr,
2131 raddr->sin_port, ep->com.cm_id->tos);
2133 ra = (__u8 *)&raddr->sin_addr;
2135 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2137 laddr6->sin6_addr.s6_addr,
2138 raddr6->sin6_addr.s6_addr,
2140 raddr6->sin6_port, 0,
2141 raddr6->sin6_scope_id);
2143 ra = (__u8 *)&raddr6->sin6_addr;
2146 pr_err("%s - cannot find route.\n", __func__);
2147 err = -EHOSTUNREACH;
2150 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2151 ep->com.dev->rdev.lldi.adapter_type,
2152 ep->com.cm_id->tos);
2154 pr_err("%s - cannot alloc l2e.\n", __func__);
2158 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2159 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2162 state_set(&ep->com, CONNECTING);
2163 ep->tos = ep->com.cm_id->tos;
2165 /* send connect request to rnic */
2166 err = send_connect(ep);
2170 cxgb4_l2t_release(ep->l2t);
2172 dst_release(ep->dst);
2174 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2175 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2178 * remember to send notification to upper layer.
2179 * We are in here so the upper layer is not aware that this is
2180 * re-connect attempt and so, upper layer is still waiting for
2181 * response of 1st connect request.
2183 connect_reply_upcall(ep, -ECONNRESET);
2185 c4iw_put_ep(&ep->com);
2190 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2193 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2194 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2195 ntohl(rpl->atid_status)));
2196 struct tid_info *t = dev->rdev.lldi.tids;
2197 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2198 struct sockaddr_in *la;
2199 struct sockaddr_in *ra;
2200 struct sockaddr_in6 *la6;
2201 struct sockaddr_in6 *ra6;
2204 ep = lookup_atid(t, atid);
2205 la = (struct sockaddr_in *)&ep->com.local_addr;
2206 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2207 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2208 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2210 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2211 status, status2errno(status));
2213 if (cxgb_is_neg_adv(status)) {
2214 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2215 __func__, atid, status, neg_adv_str(status));
2216 ep->stats.connect_neg_adv++;
2217 mutex_lock(&dev->rdev.stats.lock);
2218 dev->rdev.stats.neg_adv++;
2219 mutex_unlock(&dev->rdev.stats.lock);
2223 set_bit(ACT_OPEN_RPL, &ep->com.history);
2226 * Log interesting failures.
2229 case CPL_ERR_CONN_RESET:
2230 case CPL_ERR_CONN_TIMEDOUT:
2232 case CPL_ERR_TCAM_FULL:
2233 mutex_lock(&dev->rdev.stats.lock);
2234 dev->rdev.stats.tcam_full++;
2235 mutex_unlock(&dev->rdev.stats.lock);
2236 if (ep->com.local_addr.ss_family == AF_INET &&
2237 dev->rdev.lldi.enable_fw_ofld_conn) {
2238 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2239 ntohl(rpl->atid_status))));
2245 case CPL_ERR_CONN_EXIST:
2246 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2247 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2248 if (ep->com.remote_addr.ss_family == AF_INET6) {
2249 struct sockaddr_in6 *sin6 =
2250 (struct sockaddr_in6 *)
2251 &ep->com.local_addr;
2253 ep->com.dev->rdev.lldi.ports[0],
2255 &sin6->sin6_addr.s6_addr, 1);
2257 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2259 cxgb4_free_atid(t, atid);
2260 dst_release(ep->dst);
2261 cxgb4_l2t_release(ep->l2t);
2267 if (ep->com.local_addr.ss_family == AF_INET) {
2268 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2269 atid, status, status2errno(status),
2270 &la->sin_addr.s_addr, ntohs(la->sin_port),
2271 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2273 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2274 atid, status, status2errno(status),
2275 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2276 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2282 connect_reply_upcall(ep, status2errno(status));
2283 state_set(&ep->com, DEAD);
2285 if (ep->com.remote_addr.ss_family == AF_INET6) {
2286 struct sockaddr_in6 *sin6 =
2287 (struct sockaddr_in6 *)&ep->com.local_addr;
2288 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2289 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2291 if (status && act_open_has_tid(status))
2292 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2294 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2295 cxgb4_free_atid(t, atid);
2296 dst_release(ep->dst);
2297 cxgb4_l2t_release(ep->l2t);
2298 c4iw_put_ep(&ep->com);
2303 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2305 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2306 unsigned int stid = GET_TID(rpl);
2307 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2310 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2313 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2314 rpl->status, status2errno(rpl->status));
2315 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2316 c4iw_put_ep(&ep->com);
2321 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2323 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2324 unsigned int stid = GET_TID(rpl);
2325 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2327 PDBG("%s ep %p\n", __func__, ep);
2328 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2329 c4iw_put_ep(&ep->com);
2333 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2334 struct cpl_pass_accept_req *req)
2336 struct cpl_pass_accept_rpl *rpl;
2337 unsigned int mtu_idx;
2341 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2343 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2345 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2346 BUG_ON(skb_cloned(skb));
2350 if (!is_t4(adapter_type)) {
2351 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2353 INIT_TP_WR(rpl5, ep->hwtid);
2355 skb_trim(skb, sizeof(*rpl));
2356 INIT_TP_WR(rpl, ep->hwtid);
2358 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2361 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2362 enable_tcp_timestamps && req->tcpopt.tstamp,
2363 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2364 wscale = cxgb_compute_wscale(rcv_win);
2367 * Specify the largest window that will fit in opt0. The
2368 * remainder will be specified in the rx_data_ack.
2370 win = ep->rcv_win >> 10;
2371 if (win > RCV_BUFSIZ_M)
2373 opt0 = (nocong ? NO_CONG_F : 0) |
2376 WND_SCALE_V(wscale) |
2377 MSS_IDX_V(mtu_idx) |
2378 L2T_IDX_V(ep->l2t->idx) |
2379 TX_CHAN_V(ep->tx_chan) |
2380 SMAC_SEL_V(ep->smac_idx) |
2381 DSCP_V(ep->tos >> 2) |
2382 ULP_MODE_V(ULP_MODE_TCPDDP) |
2384 opt2 = RX_CHANNEL_V(0) |
2385 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2387 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2388 opt2 |= TSTAMPS_EN_F;
2389 if (enable_tcp_sack && req->tcpopt.sack)
2391 if (wscale && enable_tcp_window_scaling)
2392 opt2 |= WND_SCALE_EN_F;
2394 const struct tcphdr *tcph;
2395 u32 hlen = ntohl(req->hdr_len);
2397 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2398 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2401 tcph = (const void *)(req + 1) +
2402 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2403 if (tcph->ece && tcph->cwr)
2404 opt2 |= CCTRL_ECN_V(1);
2406 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2407 u32 isn = (prandom_u32() & ~7UL) - 1;
2408 opt2 |= T5_OPT_2_VALID_F;
2409 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2412 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2415 rpl5->iss = cpu_to_be32(isn);
2416 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2419 rpl->opt0 = cpu_to_be64(opt0);
2420 rpl->opt2 = cpu_to_be32(opt2);
2421 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2422 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2424 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2427 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2429 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2430 BUG_ON(skb_cloned(skb));
2431 skb_trim(skb, sizeof(struct cpl_tid_release));
2432 release_tid(&dev->rdev, hwtid, skb);
2436 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2438 struct c4iw_ep *child_ep = NULL, *parent_ep;
2439 struct cpl_pass_accept_req *req = cplhdr(skb);
2440 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2441 struct tid_info *t = dev->rdev.lldi.tids;
2442 unsigned int hwtid = GET_TID(req);
2443 struct dst_entry *dst;
2444 __u8 local_ip[16], peer_ip[16];
2445 __be16 local_port, peer_port;
2446 struct sockaddr_in6 *sin6;
2448 u16 peer_mss = ntohs(req->tcpopt.mss);
2450 unsigned short hdrs;
2451 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2453 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2455 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2459 if (state_read(&parent_ep->com) != LISTEN) {
2460 PDBG("%s - listening ep not in LISTEN\n", __func__);
2464 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2465 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2467 /* Find output route */
2469 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2470 , __func__, parent_ep, hwtid,
2471 local_ip, peer_ip, ntohs(local_port),
2472 ntohs(peer_port), peer_mss);
2473 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2474 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2475 local_port, peer_port, tos);
2477 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2478 , __func__, parent_ep, hwtid,
2479 local_ip, peer_ip, ntohs(local_port),
2480 ntohs(peer_port), peer_mss);
2481 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2482 local_ip, peer_ip, local_port, peer_port,
2483 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2484 ((struct sockaddr_in6 *)
2485 &parent_ep->com.local_addr)->sin6_scope_id);
2488 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2493 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2495 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2501 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2502 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2504 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2511 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2512 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2513 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2514 child_ep->mtu = peer_mss + hdrs;
2516 skb_queue_head_init(&child_ep->com.ep_skb_list);
2517 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2520 state_set(&child_ep->com, CONNECTING);
2521 child_ep->com.dev = dev;
2522 child_ep->com.cm_id = NULL;
2525 struct sockaddr_in *sin = (struct sockaddr_in *)
2526 &child_ep->com.local_addr;
2528 sin->sin_family = PF_INET;
2529 sin->sin_port = local_port;
2530 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2532 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2533 sin->sin_family = PF_INET;
2534 sin->sin_port = ((struct sockaddr_in *)
2535 &parent_ep->com.local_addr)->sin_port;
2536 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2538 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2539 sin->sin_family = PF_INET;
2540 sin->sin_port = peer_port;
2541 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2543 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2544 sin6->sin6_family = PF_INET6;
2545 sin6->sin6_port = local_port;
2546 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2548 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2549 sin6->sin6_family = PF_INET6;
2550 sin6->sin6_port = ((struct sockaddr_in6 *)
2551 &parent_ep->com.local_addr)->sin6_port;
2552 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2554 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2555 sin6->sin6_family = PF_INET6;
2556 sin6->sin6_port = peer_port;
2557 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2560 c4iw_get_ep(&parent_ep->com);
2561 child_ep->parent_ep = parent_ep;
2562 child_ep->tos = tos;
2563 child_ep->dst = dst;
2564 child_ep->hwtid = hwtid;
2566 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2567 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2569 init_timer(&child_ep->timer);
2570 cxgb4_insert_tid(t, child_ep, hwtid);
2571 insert_ep_tid(child_ep);
2572 if (accept_cr(child_ep, skb, req)) {
2573 c4iw_put_ep(&parent_ep->com);
2574 release_ep_resources(child_ep);
2576 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2579 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2580 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2581 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2585 c4iw_put_ep(&child_ep->com);
2587 reject_cr(dev, hwtid, skb);
2589 c4iw_put_ep(&parent_ep->com);
2594 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2597 struct cpl_pass_establish *req = cplhdr(skb);
2598 unsigned int tid = GET_TID(req);
2601 ep = get_ep_from_tid(dev, tid);
2602 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2603 ep->snd_seq = be32_to_cpu(req->snd_isn);
2604 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2606 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2607 ntohs(req->tcp_opt));
2609 set_emss(ep, ntohs(req->tcp_opt));
2611 dst_confirm(ep->dst);
2612 mutex_lock(&ep->com.mutex);
2613 ep->com.state = MPA_REQ_WAIT;
2615 set_bit(PASS_ESTAB, &ep->com.history);
2616 ret = send_flowc(ep);
2617 mutex_unlock(&ep->com.mutex);
2619 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2620 c4iw_put_ep(&ep->com);
2625 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2627 struct cpl_peer_close *hdr = cplhdr(skb);
2629 struct c4iw_qp_attributes attrs;
2632 unsigned int tid = GET_TID(hdr);
2635 ep = get_ep_from_tid(dev, tid);
2639 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2640 dst_confirm(ep->dst);
2642 set_bit(PEER_CLOSE, &ep->com.history);
2643 mutex_lock(&ep->com.mutex);
2644 switch (ep->com.state) {
2646 __state_set(&ep->com, CLOSING);
2649 __state_set(&ep->com, CLOSING);
2650 connect_reply_upcall(ep, -ECONNRESET);
2655 * We're gonna mark this puppy DEAD, but keep
2656 * the reference on it until the ULP accepts or
2657 * rejects the CR. Also wake up anyone waiting
2658 * in rdma connection migration (see c4iw_accept_cr()).
2660 __state_set(&ep->com, CLOSING);
2661 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2662 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2665 __state_set(&ep->com, CLOSING);
2666 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2667 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2671 __state_set(&ep->com, CLOSING);
2672 attrs.next_state = C4IW_QP_STATE_CLOSING;
2673 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2674 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2675 if (ret != -ECONNRESET) {
2676 peer_close_upcall(ep);
2684 __state_set(&ep->com, MORIBUND);
2688 (void)stop_ep_timer(ep);
2689 if (ep->com.cm_id && ep->com.qp) {
2690 attrs.next_state = C4IW_QP_STATE_IDLE;
2691 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2692 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2694 close_complete_upcall(ep, 0);
2695 __state_set(&ep->com, DEAD);
2705 mutex_unlock(&ep->com.mutex);
2707 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2709 release_ep_resources(ep);
2710 c4iw_put_ep(&ep->com);
2714 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2716 struct cpl_abort_req_rss *req = cplhdr(skb);
2718 struct cpl_abort_rpl *rpl;
2719 struct sk_buff *rpl_skb;
2720 struct c4iw_qp_attributes attrs;
2723 unsigned int tid = GET_TID(req);
2725 ep = get_ep_from_tid(dev, tid);
2729 if (cxgb_is_neg_adv(req->status)) {
2730 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2731 __func__, ep->hwtid, req->status,
2732 neg_adv_str(req->status));
2733 ep->stats.abort_neg_adv++;
2734 mutex_lock(&dev->rdev.stats.lock);
2735 dev->rdev.stats.neg_adv++;
2736 mutex_unlock(&dev->rdev.stats.lock);
2739 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2741 set_bit(PEER_ABORT, &ep->com.history);
2744 * Wake up any threads in rdma_init() or rdma_fini().
2745 * However, this is not needed if com state is just
2748 if (ep->com.state != MPA_REQ_SENT)
2749 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2751 mutex_lock(&ep->com.mutex);
2752 switch (ep->com.state) {
2754 c4iw_put_ep(&ep->parent_ep->com);
2757 (void)stop_ep_timer(ep);
2760 (void)stop_ep_timer(ep);
2761 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2762 connect_reply_upcall(ep, -ECONNRESET);
2765 * we just don't send notification upwards because we
2766 * want to retry with mpa_v1 without upper layers even
2769 * do some housekeeping so as to re-initiate the
2772 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2774 ep->retry_with_mpa_v1 = 1;
2786 if (ep->com.cm_id && ep->com.qp) {
2787 attrs.next_state = C4IW_QP_STATE_ERROR;
2788 ret = c4iw_modify_qp(ep->com.qp->rhp,
2789 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2793 "%s - qp <- error failed!\n",
2796 peer_abort_upcall(ep);
2801 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2802 mutex_unlock(&ep->com.mutex);
2808 dst_confirm(ep->dst);
2809 if (ep->com.state != ABORTING) {
2810 __state_set(&ep->com, DEAD);
2811 /* we don't release if we want to retry with mpa_v1 */
2812 if (!ep->retry_with_mpa_v1)
2815 mutex_unlock(&ep->com.mutex);
2817 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2818 if (WARN_ON(!rpl_skb)) {
2822 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2823 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2824 INIT_TP_WR(rpl, ep->hwtid);
2825 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2826 rpl->cmd = CPL_ABORT_NO_RST;
2827 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2830 release_ep_resources(ep);
2831 else if (ep->retry_with_mpa_v1) {
2832 if (ep->com.remote_addr.ss_family == AF_INET6) {
2833 struct sockaddr_in6 *sin6 =
2834 (struct sockaddr_in6 *)
2835 &ep->com.local_addr;
2837 ep->com.dev->rdev.lldi.ports[0],
2838 (const u32 *)&sin6->sin6_addr.s6_addr,
2841 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2842 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2843 dst_release(ep->dst);
2844 cxgb4_l2t_release(ep->l2t);
2849 c4iw_put_ep(&ep->com);
2850 /* Dereferencing ep, referenced in peer_abort_intr() */
2851 c4iw_put_ep(&ep->com);
2855 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2858 struct c4iw_qp_attributes attrs;
2859 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2861 unsigned int tid = GET_TID(rpl);
2863 ep = get_ep_from_tid(dev, tid);
2867 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2870 /* The cm_id may be null if we failed to connect */
2871 mutex_lock(&ep->com.mutex);
2872 set_bit(CLOSE_CON_RPL, &ep->com.history);
2873 switch (ep->com.state) {
2875 __state_set(&ep->com, MORIBUND);
2878 (void)stop_ep_timer(ep);
2879 if ((ep->com.cm_id) && (ep->com.qp)) {
2880 attrs.next_state = C4IW_QP_STATE_IDLE;
2881 c4iw_modify_qp(ep->com.qp->rhp,
2883 C4IW_QP_ATTR_NEXT_STATE,
2886 close_complete_upcall(ep, 0);
2887 __state_set(&ep->com, DEAD);
2897 mutex_unlock(&ep->com.mutex);
2899 release_ep_resources(ep);
2900 c4iw_put_ep(&ep->com);
2904 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2906 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2907 unsigned int tid = GET_TID(rpl);
2909 struct c4iw_qp_attributes attrs;
2911 ep = get_ep_from_tid(dev, tid);
2914 if (ep && ep->com.qp) {
2915 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2916 ep->com.qp->wq.sq.qid);
2917 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2918 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2919 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2921 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2922 c4iw_put_ep(&ep->com);
2928 * Upcall from the adapter indicating data has been transmitted.
2929 * For us its just the single MPA request or reply. We can now free
2930 * the skb holding the mpa message.
2932 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2935 struct cpl_fw4_ack *hdr = cplhdr(skb);
2936 u8 credits = hdr->credits;
2937 unsigned int tid = GET_TID(hdr);
2940 ep = get_ep_from_tid(dev, tid);
2943 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2945 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2946 __func__, ep, ep->hwtid, state_read(&ep->com));
2950 dst_confirm(ep->dst);
2952 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2953 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2954 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2955 mutex_lock(&ep->com.mutex);
2956 kfree_skb(ep->mpa_skb);
2958 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
2960 mutex_unlock(&ep->com.mutex);
2963 c4iw_put_ep(&ep->com);
2967 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2970 struct c4iw_ep *ep = to_ep(cm_id);
2972 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2974 mutex_lock(&ep->com.mutex);
2975 if (ep->com.state != MPA_REQ_RCVD) {
2976 mutex_unlock(&ep->com.mutex);
2977 c4iw_put_ep(&ep->com);
2980 set_bit(ULP_REJECT, &ep->com.history);
2984 abort = send_mpa_reject(ep, pdata, pdata_len);
2985 mutex_unlock(&ep->com.mutex);
2988 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
2989 c4iw_put_ep(&ep->com);
2993 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2996 struct c4iw_qp_attributes attrs;
2997 enum c4iw_qp_attr_mask mask;
2998 struct c4iw_ep *ep = to_ep(cm_id);
2999 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3000 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3003 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3005 mutex_lock(&ep->com.mutex);
3006 if (ep->com.state != MPA_REQ_RCVD) {
3013 set_bit(ULP_ACCEPT, &ep->com.history);
3014 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3015 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3020 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3021 if (conn_param->ord > ep->ird) {
3022 if (RELAXED_IRD_NEGOTIATION) {
3023 conn_param->ord = ep->ird;
3025 ep->ird = conn_param->ird;
3026 ep->ord = conn_param->ord;
3027 send_mpa_reject(ep, conn_param->private_data,
3028 conn_param->private_data_len);
3033 if (conn_param->ird < ep->ord) {
3034 if (RELAXED_IRD_NEGOTIATION &&
3035 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3036 conn_param->ird = ep->ord;
3043 ep->ird = conn_param->ird;
3044 ep->ord = conn_param->ord;
3046 if (ep->mpa_attr.version == 1) {
3047 if (peer2peer && ep->ird == 0)
3051 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3052 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3056 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3058 ep->com.cm_id = cm_id;
3059 ref_cm_id(&ep->com);
3063 /* bind QP to EP and move to RTS */
3064 attrs.mpa_attr = ep->mpa_attr;
3065 attrs.max_ird = ep->ird;
3066 attrs.max_ord = ep->ord;
3067 attrs.llp_stream_handle = ep;
3068 attrs.next_state = C4IW_QP_STATE_RTS;
3070 /* bind QP and TID with INIT_WR */
3071 mask = C4IW_QP_ATTR_NEXT_STATE |
3072 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3073 C4IW_QP_ATTR_MPA_ATTR |
3074 C4IW_QP_ATTR_MAX_IRD |
3075 C4IW_QP_ATTR_MAX_ORD;
3077 err = c4iw_modify_qp(ep->com.qp->rhp,
3078 ep->com.qp, mask, &attrs, 1);
3080 goto err_deref_cm_id;
3082 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3083 err = send_mpa_reply(ep, conn_param->private_data,
3084 conn_param->private_data_len);
3086 goto err_deref_cm_id;
3088 __state_set(&ep->com, FPDU_MODE);
3089 established_upcall(ep);
3090 mutex_unlock(&ep->com.mutex);
3091 c4iw_put_ep(&ep->com);
3094 deref_cm_id(&ep->com);
3098 mutex_unlock(&ep->com.mutex);
3100 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3101 c4iw_put_ep(&ep->com);
3105 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3107 struct in_device *ind;
3109 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3110 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3112 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3114 return -EADDRNOTAVAIL;
3115 for_primary_ifa(ind) {
3116 laddr->sin_addr.s_addr = ifa->ifa_address;
3117 raddr->sin_addr.s_addr = ifa->ifa_address;
3123 return found ? 0 : -EADDRNOTAVAIL;
3126 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3127 unsigned char banned_flags)
3129 struct inet6_dev *idev;
3130 int err = -EADDRNOTAVAIL;
3133 idev = __in6_dev_get(dev);
3135 struct inet6_ifaddr *ifp;
3137 read_lock_bh(&idev->lock);
3138 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3139 if (ifp->scope == IFA_LINK &&
3140 !(ifp->flags & banned_flags)) {
3141 memcpy(addr, &ifp->addr, 16);
3146 read_unlock_bh(&idev->lock);
3152 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3154 struct in6_addr uninitialized_var(addr);
3155 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3156 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3158 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3159 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3160 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3163 return -EADDRNOTAVAIL;
3166 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3168 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3171 struct sockaddr_in *laddr;
3172 struct sockaddr_in *raddr;
3173 struct sockaddr_in6 *laddr6;
3174 struct sockaddr_in6 *raddr6;
3178 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3179 (conn_param->ird > cur_max_read_depth(dev))) {
3183 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3185 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3190 skb_queue_head_init(&ep->com.ep_skb_list);
3191 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3196 init_timer(&ep->timer);
3197 ep->plen = conn_param->private_data_len;
3199 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3200 conn_param->private_data, ep->plen);
3201 ep->ird = conn_param->ird;
3202 ep->ord = conn_param->ord;
3204 if (peer2peer && ep->ord == 0)
3207 ep->com.cm_id = cm_id;
3208 ref_cm_id(&ep->com);
3210 ep->com.qp = get_qhp(dev, conn_param->qpn);
3212 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3217 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3221 * Allocate an active TID to initiate a TCP connection.
3223 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3224 if (ep->atid == -1) {
3225 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3229 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3231 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3232 sizeof(ep->com.local_addr));
3233 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3234 sizeof(ep->com.remote_addr));
3236 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3237 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3238 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3239 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3241 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3243 ra = (__u8 *)&raddr->sin_addr;
3246 * Handle loopback requests to INADDR_ANY.
3248 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3249 err = pick_local_ipaddrs(dev, cm_id);
3255 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3256 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3257 ra, ntohs(raddr->sin_port));
3258 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3259 laddr->sin_addr.s_addr,
3260 raddr->sin_addr.s_addr,
3262 raddr->sin_port, cm_id->tos);
3265 ra = (__u8 *)&raddr6->sin6_addr;
3268 * Handle loopback requests to INADDR_ANY.
3270 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3271 err = pick_local_ip6addrs(dev, cm_id);
3277 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3278 __func__, laddr6->sin6_addr.s6_addr,
3279 ntohs(laddr6->sin6_port),
3280 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3281 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3282 laddr6->sin6_addr.s6_addr,
3283 raddr6->sin6_addr.s6_addr,
3285 raddr6->sin6_port, 0,
3286 raddr6->sin6_scope_id);
3289 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3290 err = -EHOSTUNREACH;
3294 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3295 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3297 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3301 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3302 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3305 state_set(&ep->com, CONNECTING);
3306 ep->tos = cm_id->tos;
3308 /* send connect request to rnic */
3309 err = send_connect(ep);
3313 cxgb4_l2t_release(ep->l2t);
3315 dst_release(ep->dst);
3317 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3318 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3320 skb_queue_purge(&ep->com.ep_skb_list);
3321 deref_cm_id(&ep->com);
3323 c4iw_put_ep(&ep->com);
3328 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3331 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3332 &ep->com.local_addr;
3334 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3335 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3336 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3340 c4iw_init_wr_wait(&ep->com.wr_wait);
3341 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3342 ep->stid, &sin6->sin6_addr,
3344 ep->com.dev->rdev.lldi.rxq_ids[0]);
3346 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3350 err = net_xmit_errno(err);
3352 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3353 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3354 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3356 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3361 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3364 struct sockaddr_in *sin = (struct sockaddr_in *)
3365 &ep->com.local_addr;
3367 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3369 err = cxgb4_create_server_filter(
3370 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3371 sin->sin_addr.s_addr, sin->sin_port, 0,
3372 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3373 if (err == -EBUSY) {
3374 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3378 set_current_state(TASK_UNINTERRUPTIBLE);
3379 schedule_timeout(usecs_to_jiffies(100));
3381 } while (err == -EBUSY);
3383 c4iw_init_wr_wait(&ep->com.wr_wait);
3384 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3385 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3386 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3388 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3392 err = net_xmit_errno(err);
3395 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3397 &sin->sin_addr, ntohs(sin->sin_port));
3401 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3404 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3405 struct c4iw_listen_ep *ep;
3409 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3411 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3415 skb_queue_head_init(&ep->com.ep_skb_list);
3416 PDBG("%s ep %p\n", __func__, ep);
3417 ep->com.cm_id = cm_id;
3418 ref_cm_id(&ep->com);
3420 ep->backlog = backlog;
3421 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3422 sizeof(ep->com.local_addr));
3425 * Allocate a server TID.
3427 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3428 ep->com.local_addr.ss_family == AF_INET)
3429 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3430 cm_id->m_local_addr.ss_family, ep);
3432 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3433 cm_id->m_local_addr.ss_family, ep);
3435 if (ep->stid == -1) {
3436 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3440 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3442 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3443 sizeof(ep->com.local_addr));
3445 state_set(&ep->com, LISTEN);
3446 if (ep->com.local_addr.ss_family == AF_INET)
3447 err = create_server4(dev, ep);
3449 err = create_server6(dev, ep);
3451 cm_id->provider_data = ep;
3455 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3456 ep->com.local_addr.ss_family);
3458 deref_cm_id(&ep->com);
3459 c4iw_put_ep(&ep->com);
3465 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3468 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3470 PDBG("%s ep %p\n", __func__, ep);
3473 state_set(&ep->com, DEAD);
3474 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3475 ep->com.local_addr.ss_family == AF_INET) {
3476 err = cxgb4_remove_server_filter(
3477 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3478 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3480 struct sockaddr_in6 *sin6;
3481 c4iw_init_wr_wait(&ep->com.wr_wait);
3482 err = cxgb4_remove_server(
3483 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3484 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3487 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3489 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3490 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3491 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3493 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3494 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3495 ep->com.local_addr.ss_family);
3497 deref_cm_id(&ep->com);
3498 c4iw_put_ep(&ep->com);
3502 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3507 struct c4iw_rdev *rdev;
3509 mutex_lock(&ep->com.mutex);
3511 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3512 states[ep->com.state], abrupt);
3515 * Ref the ep here in case we have fatal errors causing the
3516 * ep to be released and freed.
3518 c4iw_get_ep(&ep->com);
3520 rdev = &ep->com.dev->rdev;
3521 if (c4iw_fatal_error(rdev)) {
3523 close_complete_upcall(ep, -EIO);
3524 ep->com.state = DEAD;
3526 switch (ep->com.state) {
3535 ep->com.state = ABORTING;
3537 ep->com.state = CLOSING;
3540 * if we close before we see the fw4_ack() then we fix
3541 * up the timer state since we're reusing it.
3544 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3545 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3550 set_bit(CLOSE_SENT, &ep->com.flags);
3553 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3556 (void)stop_ep_timer(ep);
3557 ep->com.state = ABORTING;
3559 ep->com.state = MORIBUND;
3565 PDBG("%s ignoring disconnect ep %p state %u\n",
3566 __func__, ep, ep->com.state);
3575 set_bit(EP_DISC_ABORT, &ep->com.history);
3576 close_complete_upcall(ep, -ECONNRESET);
3577 ret = send_abort(ep);
3579 set_bit(EP_DISC_CLOSE, &ep->com.history);
3580 ret = send_halfclose(ep);
3583 set_bit(EP_DISC_FAIL, &ep->com.history);
3586 close_complete_upcall(ep, -EIO);
3589 struct c4iw_qp_attributes attrs;
3591 attrs.next_state = C4IW_QP_STATE_ERROR;
3592 ret = c4iw_modify_qp(ep->com.qp->rhp,
3594 C4IW_QP_ATTR_NEXT_STATE,
3598 "%s - qp <- error failed!\n",
3604 mutex_unlock(&ep->com.mutex);
3605 c4iw_put_ep(&ep->com);
3607 release_ep_resources(ep);
3611 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3612 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3615 int atid = be32_to_cpu(req->tid);
3617 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3618 (__force u32) req->tid);
3622 switch (req->retval) {
3624 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3625 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3626 send_fw_act_open_req(ep, atid);
3630 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3631 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3632 send_fw_act_open_req(ep, atid);
3637 pr_info("%s unexpected ofld conn wr retval %d\n",
3638 __func__, req->retval);
3641 pr_err("active ofld_connect_wr failure %d atid %d\n",
3643 mutex_lock(&dev->rdev.stats.lock);
3644 dev->rdev.stats.act_ofld_conn_fails++;
3645 mutex_unlock(&dev->rdev.stats.lock);
3646 connect_reply_upcall(ep, status2errno(req->retval));
3647 state_set(&ep->com, DEAD);
3648 if (ep->com.remote_addr.ss_family == AF_INET6) {
3649 struct sockaddr_in6 *sin6 =
3650 (struct sockaddr_in6 *)&ep->com.local_addr;
3651 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3652 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3654 remove_handle(dev, &dev->atid_idr, atid);
3655 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3656 dst_release(ep->dst);
3657 cxgb4_l2t_release(ep->l2t);
3658 c4iw_put_ep(&ep->com);
3661 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3662 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3664 struct sk_buff *rpl_skb;
3665 struct cpl_pass_accept_req *cpl;
3668 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3671 PDBG("%s passive open failure %d\n", __func__, req->retval);
3672 mutex_lock(&dev->rdev.stats.lock);
3673 dev->rdev.stats.pas_ofld_conn_fails++;
3674 mutex_unlock(&dev->rdev.stats.lock);
3677 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3678 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3679 (__force u32) htonl(
3680 (__force u32) req->tid)));
3681 ret = pass_accept_req(dev, rpl_skb);
3688 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3690 struct cpl_fw6_msg *rpl = cplhdr(skb);
3691 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3693 switch (rpl->type) {
3695 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3697 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3698 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3699 switch (req->t_state) {
3701 active_ofld_conn_reply(dev, skb, req);
3704 passive_ofld_conn_reply(dev, skb, req);
3707 pr_err("%s unexpected ofld conn wr state %d\n",
3708 __func__, req->t_state);
3716 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3719 __be16 hdr_len, vlantag, len;
3721 int tcp_hdr_len, ip_hdr_len;
3723 struct cpl_rx_pkt *cpl = cplhdr(skb);
3724 struct cpl_pass_accept_req *req;
3725 struct tcp_options_received tmp_opt;
3726 struct c4iw_dev *dev;
3727 enum chip_type type;
3729 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3730 /* Store values from cpl_rx_pkt in temporary location. */
3731 vlantag = cpl->vlan;
3733 l2info = cpl->l2info;
3734 hdr_len = cpl->hdr_len;
3737 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3740 * We need to parse the TCP options from SYN packet.
3741 * to generate cpl_pass_accept_req.
3743 memset(&tmp_opt, 0, sizeof(tmp_opt));
3744 tcp_clear_options(&tmp_opt);
3745 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3747 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3748 memset(req, 0, sizeof(*req));
3749 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3750 SYN_MAC_IDX_V(RX_MACIDX_G(
3751 be32_to_cpu(l2info))) |
3753 type = dev->rdev.lldi.adapter_type;
3754 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3755 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3757 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3758 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3759 eth_hdr_len = is_t4(type) ?
3760 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3761 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3762 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3763 IP_HDR_LEN_V(ip_hdr_len) |
3764 ETH_HDR_LEN_V(eth_hdr_len));
3765 } else { /* T6 and later */
3766 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3767 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3768 T6_IP_HDR_LEN_V(ip_hdr_len) |
3769 T6_ETH_HDR_LEN_V(eth_hdr_len));
3771 req->vlan = vlantag;
3773 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3774 PASS_OPEN_TOS_V(tos));
3775 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3776 if (tmp_opt.wscale_ok)
3777 req->tcpopt.wsf = tmp_opt.snd_wscale;
3778 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3779 if (tmp_opt.sack_ok)
3780 req->tcpopt.sack = 1;
3781 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3785 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3786 __be32 laddr, __be16 lport,
3787 __be32 raddr, __be16 rport,
3788 u32 rcv_isn, u32 filter, u16 window,
3789 u32 rss_qid, u8 port_id)
3791 struct sk_buff *req_skb;
3792 struct fw_ofld_connection_wr *req;
3793 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3796 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3797 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3798 memset(req, 0, sizeof(*req));
3799 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3800 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3801 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3802 req->le.filter = (__force __be32) filter;
3803 req->le.lport = lport;
3804 req->le.pport = rport;
3805 req->le.u.ipv4.lip = laddr;
3806 req->le.u.ipv4.pip = raddr;
3807 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3808 req->tcb.rcv_adv = htons(window);
3809 req->tcb.t_state_to_astid =
3810 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3811 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3812 FW_OFLD_CONNECTION_WR_ASTID_V(
3813 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3816 * We store the qid in opt2 which will be used by the firmware
3817 * to send us the wr response.
3819 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3822 * We initialize the MSS index in TCB to 0xF.
3823 * So that when driver sends cpl_pass_accept_rpl
3824 * TCB picks up the correct value. If this was 0
3825 * TP will ignore any value > 0 for MSS index.
3827 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3828 req->cookie = (uintptr_t)skb;
3830 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3831 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3833 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3841 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3842 * messages when a filter is being used instead of server to
3843 * redirect a syn packet. When packets hit filter they are redirected
3844 * to the offload queue and driver tries to establish the connection
3845 * using firmware work request.
3847 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3850 unsigned int filter;
3851 struct ethhdr *eh = NULL;
3852 struct vlan_ethhdr *vlan_eh = NULL;
3854 struct tcphdr *tcph;
3855 struct rss_header *rss = (void *)skb->data;
3856 struct cpl_rx_pkt *cpl = (void *)skb->data;
3857 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3858 struct l2t_entry *e;
3859 struct dst_entry *dst;
3860 struct c4iw_ep *lep = NULL;
3862 struct port_info *pi;
3863 struct net_device *pdev;
3864 u16 rss_qid, eth_hdr_len;
3867 struct neighbour *neigh;
3869 /* Drop all non-SYN packets */
3870 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3874 * Drop all packets which did not hit the filter.
3875 * Unlikely to happen.
3877 if (!(rss->filter_hit && rss->filter_tid))
3881 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3883 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3885 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3887 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3891 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3893 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3896 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3899 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3902 pr_err("T%d Chip is not supported\n",
3903 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3907 if (eth_hdr_len == ETH_HLEN) {
3908 eh = (struct ethhdr *)(req + 1);
3909 iph = (struct iphdr *)(eh + 1);
3911 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3912 iph = (struct iphdr *)(vlan_eh + 1);
3913 skb->vlan_tci = ntohs(cpl->vlan);
3916 if (iph->version != 0x4)
3919 tcph = (struct tcphdr *)(iph + 1);
3920 skb_set_network_header(skb, (void *)iph - (void *)rss);
3921 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3924 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3925 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3926 ntohs(tcph->source), iph->tos);
3928 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3929 iph->daddr, iph->saddr, tcph->dest,
3930 tcph->source, iph->tos);
3932 pr_err("%s - failed to find dst entry!\n",
3936 neigh = dst_neigh_lookup_skb(dst, skb);
3939 pr_err("%s - failed to allocate neigh!\n",
3944 if (neigh->dev->flags & IFF_LOOPBACK) {
3945 pdev = ip_dev_find(&init_net, iph->daddr);
3946 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3948 pi = (struct port_info *)netdev_priv(pdev);
3949 tx_chan = cxgb4_port_chan(pdev);
3952 pdev = get_real_dev(neigh->dev);
3953 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3955 pi = (struct port_info *)netdev_priv(pdev);
3956 tx_chan = cxgb4_port_chan(pdev);
3958 neigh_release(neigh);
3960 pr_err("%s - failed to allocate l2t entry!\n",
3965 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3966 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3967 window = (__force u16) htons((__force u16)tcph->window);
3969 /* Calcuate filter portion for LE region. */
3970 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3971 dev->rdev.lldi.ports[0],
3975 * Synthesize the cpl_pass_accept_req. We have everything except the
3976 * TID. Once firmware sends a reply with TID we update the TID field
3977 * in cpl and pass it through the regular cpl_pass_accept_req path.
3979 build_cpl_pass_accept_req(skb, stid, iph->tos);
3980 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3981 tcph->source, ntohl(tcph->seq), filter, window,
3982 rss_qid, pi->port_id);
3983 cxgb4_l2t_release(e);
3988 c4iw_put_ep(&lep->com);
3993 * These are the real handlers that are called from a
3996 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
3997 [CPL_ACT_ESTABLISH] = act_establish,
3998 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3999 [CPL_RX_DATA] = rx_data,
4000 [CPL_ABORT_RPL_RSS] = abort_rpl,
4001 [CPL_ABORT_RPL] = abort_rpl,
4002 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4003 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4004 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4005 [CPL_PASS_ESTABLISH] = pass_establish,
4006 [CPL_PEER_CLOSE] = peer_close,
4007 [CPL_ABORT_REQ_RSS] = peer_abort,
4008 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4009 [CPL_RDMA_TERMINATE] = terminate,
4010 [CPL_FW4_ACK] = fw4_ack,
4011 [CPL_FW6_MSG] = deferred_fw6_msg,
4012 [CPL_RX_PKT] = rx_pkt,
4013 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4014 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4017 static void process_timeout(struct c4iw_ep *ep)
4019 struct c4iw_qp_attributes attrs;
4022 mutex_lock(&ep->com.mutex);
4023 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4025 set_bit(TIMEDOUT, &ep->com.history);
4026 switch (ep->com.state) {
4028 connect_reply_upcall(ep, -ETIMEDOUT);
4037 if (ep->com.cm_id && ep->com.qp) {
4038 attrs.next_state = C4IW_QP_STATE_ERROR;
4039 c4iw_modify_qp(ep->com.qp->rhp,
4040 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4043 close_complete_upcall(ep, -ETIMEDOUT);
4049 * These states are expected if the ep timed out at the same
4050 * time as another thread was calling stop_ep_timer().
4051 * So we silently do nothing for these states.
4056 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4057 __func__, ep, ep->hwtid, ep->com.state);
4060 mutex_unlock(&ep->com.mutex);
4062 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4063 c4iw_put_ep(&ep->com);
4066 static void process_timedout_eps(void)
4070 spin_lock_irq(&timeout_lock);
4071 while (!list_empty(&timeout_list)) {
4072 struct list_head *tmp;
4074 tmp = timeout_list.next;
4078 spin_unlock_irq(&timeout_lock);
4079 ep = list_entry(tmp, struct c4iw_ep, entry);
4080 process_timeout(ep);
4081 spin_lock_irq(&timeout_lock);
4083 spin_unlock_irq(&timeout_lock);
4086 static void process_work(struct work_struct *work)
4088 struct sk_buff *skb = NULL;
4089 struct c4iw_dev *dev;
4090 struct cpl_act_establish *rpl;
4091 unsigned int opcode;
4094 process_timedout_eps();
4095 while ((skb = skb_dequeue(&rxq))) {
4097 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4098 opcode = rpl->ot.opcode;
4100 BUG_ON(!work_handlers[opcode]);
4101 ret = work_handlers[opcode](dev, skb);
4104 process_timedout_eps();
4108 static DECLARE_WORK(skb_work, process_work);
4110 static void ep_timeout(unsigned long arg)
4112 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4115 spin_lock(&timeout_lock);
4116 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4118 * Only insert if it is not already on the list.
4120 if (!ep->entry.next) {
4121 list_add_tail(&ep->entry, &timeout_list);
4125 spin_unlock(&timeout_lock);
4127 queue_work(workq, &skb_work);
4131 * All the CM events are handled on a work queue to have a safe context.
4133 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4137 * Save dev in the skb->cb area.
4139 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4142 * Queue the skb and schedule the worker thread.
4144 skb_queue_tail(&rxq, skb);
4145 queue_work(workq, &skb_work);
4149 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4151 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4153 if (rpl->status != CPL_ERR_NONE) {
4154 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4155 "for tid %u\n", rpl->status, GET_TID(rpl));
4161 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4163 struct cpl_fw6_msg *rpl = cplhdr(skb);
4164 struct c4iw_wr_wait *wr_waitp;
4167 PDBG("%s type %u\n", __func__, rpl->type);
4169 switch (rpl->type) {
4170 case FW6_TYPE_WR_RPL:
4171 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4172 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4173 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4175 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4179 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4183 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4191 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4193 struct cpl_abort_req_rss *req = cplhdr(skb);
4195 unsigned int tid = GET_TID(req);
4197 ep = get_ep_from_tid(dev, tid);
4198 /* This EP will be dereferenced in peer_abort() */
4200 printk(KERN_WARNING MOD
4201 "Abort on non-existent endpoint, tid %d\n", tid);
4205 if (cxgb_is_neg_adv(req->status)) {
4206 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4207 __func__, ep->hwtid, req->status,
4208 neg_adv_str(req->status));
4211 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4214 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4221 * Most upcalls from the T4 Core go to sched() to
4222 * schedule the processing on a work queue.
4224 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4225 [CPL_ACT_ESTABLISH] = sched,
4226 [CPL_ACT_OPEN_RPL] = sched,
4227 [CPL_RX_DATA] = sched,
4228 [CPL_ABORT_RPL_RSS] = sched,
4229 [CPL_ABORT_RPL] = sched,
4230 [CPL_PASS_OPEN_RPL] = sched,
4231 [CPL_CLOSE_LISTSRV_RPL] = sched,
4232 [CPL_PASS_ACCEPT_REQ] = sched,
4233 [CPL_PASS_ESTABLISH] = sched,
4234 [CPL_PEER_CLOSE] = sched,
4235 [CPL_CLOSE_CON_RPL] = sched,
4236 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4237 [CPL_RDMA_TERMINATE] = sched,
4238 [CPL_FW4_ACK] = sched,
4239 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4240 [CPL_FW6_MSG] = fw6_msg,
4241 [CPL_RX_PKT] = sched
4244 int __init c4iw_cm_init(void)
4246 spin_lock_init(&timeout_lock);
4247 skb_queue_head_init(&rxq);
4249 workq = create_singlethread_workqueue("iw_cxgb4");
4256 void c4iw_cm_term(void)
4258 WARN_ON(!list_empty(&timeout_list));
4259 flush_workqueue(workq);
4260 destroy_workqueue(workq);