Merge tag 'block-5.11-2021-01-10' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / drivers / infiniband / hw / cxgb4 / cm.c
blob8769e7aa097f4f6c009ee14f2793e3e9330327e8
1 /*
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
12 * conditions are met:
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
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
30 * SOFTWARE.
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>
39 #include <linux/ip.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>
46 #include <net/tcp.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
50 #include <rdma/ib_addr.h>
52 #include <libcxgb_cm.h>
53 #include "iw_cxgb4.h"
54 #include "clip_tbl.h"
56 static char *states[] = {
57 "idle",
58 "listen",
59 "connecting",
60 "mpa_wait_req",
61 "mpa_req_sent",
62 "mpa_req_rcvd",
63 "mpa_rep_sent",
64 "fpdu_mode",
65 "aborting",
66 "closing",
67 "moribund",
68 "dead",
69 NULL,
72 static int nocong;
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;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=0)");
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)");
102 static int peer2peer = 1;
103 module_param(peer2peer, int, 0644);
104 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
106 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
107 module_param(p2p_type, int, 0644);
108 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
109 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
111 static int ep_timeout_secs = 60;
112 module_param(ep_timeout_secs, int, 0644);
113 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
114 "in seconds (default=60)");
116 static int mpa_rev = 2;
117 module_param(mpa_rev, int, 0644);
118 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
119 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
120 " compliant (default=2)");
122 static int markers_enabled;
123 module_param(markers_enabled, int, 0644);
124 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
126 static int crc_enabled = 1;
127 module_param(crc_enabled, int, 0644);
128 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
130 static int rcv_win = 256 * 1024;
131 module_param(rcv_win, int, 0644);
132 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
134 static int snd_win = 128 * 1024;
135 module_param(snd_win, int, 0644);
136 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
138 static struct workqueue_struct *workq;
140 static struct sk_buff_head rxq;
142 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
143 static void ep_timeout(struct timer_list *t);
144 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
145 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
147 static LIST_HEAD(timeout_list);
148 static spinlock_t timeout_lock;
150 static void deref_cm_id(struct c4iw_ep_common *epc)
152 epc->cm_id->rem_ref(epc->cm_id);
153 epc->cm_id = NULL;
154 set_bit(CM_ID_DEREFED, &epc->history);
157 static void ref_cm_id(struct c4iw_ep_common *epc)
159 set_bit(CM_ID_REFED, &epc->history);
160 epc->cm_id->add_ref(epc->cm_id);
163 static void deref_qp(struct c4iw_ep *ep)
165 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
166 clear_bit(QP_REFERENCED, &ep->com.flags);
167 set_bit(QP_DEREFED, &ep->com.history);
170 static void ref_qp(struct c4iw_ep *ep)
172 set_bit(QP_REFERENCED, &ep->com.flags);
173 set_bit(QP_REFED, &ep->com.history);
174 c4iw_qp_add_ref(&ep->com.qp->ibqp);
177 static void start_ep_timer(struct c4iw_ep *ep)
179 pr_debug("ep %p\n", ep);
180 if (timer_pending(&ep->timer)) {
181 pr_err("%s timer already started! ep %p\n",
182 __func__, ep);
183 return;
185 clear_bit(TIMEOUT, &ep->com.flags);
186 c4iw_get_ep(&ep->com);
187 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
188 add_timer(&ep->timer);
191 static int stop_ep_timer(struct c4iw_ep *ep)
193 pr_debug("ep %p stopping\n", ep);
194 del_timer_sync(&ep->timer);
195 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
196 c4iw_put_ep(&ep->com);
197 return 0;
199 return 1;
202 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
203 struct l2t_entry *l2e)
205 int error = 0;
207 if (c4iw_fatal_error(rdev)) {
208 kfree_skb(skb);
209 pr_err("%s - device in error state - dropping\n", __func__);
210 return -EIO;
212 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
213 if (error < 0)
214 kfree_skb(skb);
215 else if (error == NET_XMIT_DROP)
216 return -ENOMEM;
217 return error < 0 ? error : 0;
220 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
222 int error = 0;
224 if (c4iw_fatal_error(rdev)) {
225 kfree_skb(skb);
226 pr_err("%s - device in error state - dropping\n", __func__);
227 return -EIO;
229 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
230 if (error < 0)
231 kfree_skb(skb);
232 return error < 0 ? error : 0;
235 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
237 u32 len = roundup(sizeof(struct cpl_tid_release), 16);
239 skb = get_skb(skb, len, GFP_KERNEL);
240 if (!skb)
241 return;
243 cxgb_mk_tid_release(skb, len, hwtid, 0);
244 c4iw_ofld_send(rdev, skb);
245 return;
248 static void set_emss(struct c4iw_ep *ep, u16 opt)
250 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
251 ((AF_INET == ep->com.remote_addr.ss_family) ?
252 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
253 sizeof(struct tcphdr);
254 ep->mss = ep->emss;
255 if (TCPOPT_TSTAMP_G(opt))
256 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
257 if (ep->emss < 128)
258 ep->emss = 128;
259 if (ep->emss & 7)
260 pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n",
261 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
262 pr_debug("mss_idx %u mss %u emss=%u\n", TCPOPT_MSS_G(opt), ep->mss,
263 ep->emss);
266 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
268 enum c4iw_ep_state state;
270 mutex_lock(&epc->mutex);
271 state = epc->state;
272 mutex_unlock(&epc->mutex);
273 return state;
276 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
278 epc->state = new;
281 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
283 mutex_lock(&epc->mutex);
284 pr_debug("%s -> %s\n", states[epc->state], states[new]);
285 __state_set(epc, new);
286 mutex_unlock(&epc->mutex);
287 return;
290 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
292 struct sk_buff *skb;
293 unsigned int i;
294 size_t len;
296 len = roundup(sizeof(union cpl_wr_size), 16);
297 for (i = 0; i < size; i++) {
298 skb = alloc_skb(len, GFP_KERNEL);
299 if (!skb)
300 goto fail;
301 skb_queue_tail(ep_skb_list, skb);
303 return 0;
304 fail:
305 skb_queue_purge(ep_skb_list);
306 return -ENOMEM;
309 static void *alloc_ep(int size, gfp_t gfp)
311 struct c4iw_ep_common *epc;
313 epc = kzalloc(size, gfp);
314 if (epc) {
315 epc->wr_waitp = c4iw_alloc_wr_wait(gfp);
316 if (!epc->wr_waitp) {
317 kfree(epc);
318 epc = NULL;
319 goto out;
321 kref_init(&epc->kref);
322 mutex_init(&epc->mutex);
323 c4iw_init_wr_wait(epc->wr_waitp);
325 pr_debug("alloc ep %p\n", epc);
326 out:
327 return epc;
330 static void remove_ep_tid(struct c4iw_ep *ep)
332 unsigned long flags;
334 xa_lock_irqsave(&ep->com.dev->hwtids, flags);
335 __xa_erase(&ep->com.dev->hwtids, ep->hwtid);
336 if (xa_empty(&ep->com.dev->hwtids))
337 wake_up(&ep->com.dev->wait);
338 xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
341 static int insert_ep_tid(struct c4iw_ep *ep)
343 unsigned long flags;
344 int err;
346 xa_lock_irqsave(&ep->com.dev->hwtids, flags);
347 err = __xa_insert(&ep->com.dev->hwtids, ep->hwtid, ep, GFP_KERNEL);
348 xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
350 return err;
354 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
356 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
358 struct c4iw_ep *ep;
359 unsigned long flags;
361 xa_lock_irqsave(&dev->hwtids, flags);
362 ep = xa_load(&dev->hwtids, tid);
363 if (ep)
364 c4iw_get_ep(&ep->com);
365 xa_unlock_irqrestore(&dev->hwtids, flags);
366 return ep;
370 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
372 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
373 unsigned int stid)
375 struct c4iw_listen_ep *ep;
376 unsigned long flags;
378 xa_lock_irqsave(&dev->stids, flags);
379 ep = xa_load(&dev->stids, stid);
380 if (ep)
381 c4iw_get_ep(&ep->com);
382 xa_unlock_irqrestore(&dev->stids, flags);
383 return ep;
386 void _c4iw_free_ep(struct kref *kref)
388 struct c4iw_ep *ep;
390 ep = container_of(kref, struct c4iw_ep, com.kref);
391 pr_debug("ep %p state %s\n", ep, states[ep->com.state]);
392 if (test_bit(QP_REFERENCED, &ep->com.flags))
393 deref_qp(ep);
394 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
395 if (ep->com.remote_addr.ss_family == AF_INET6) {
396 struct sockaddr_in6 *sin6 =
397 (struct sockaddr_in6 *)
398 &ep->com.local_addr;
400 cxgb4_clip_release(
401 ep->com.dev->rdev.lldi.ports[0],
402 (const u32 *)&sin6->sin6_addr.s6_addr,
405 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
406 ep->com.local_addr.ss_family);
407 dst_release(ep->dst);
408 cxgb4_l2t_release(ep->l2t);
409 kfree_skb(ep->mpa_skb);
411 if (!skb_queue_empty(&ep->com.ep_skb_list))
412 skb_queue_purge(&ep->com.ep_skb_list);
413 c4iw_put_wr_wait(ep->com.wr_waitp);
414 kfree(ep);
417 static void release_ep_resources(struct c4iw_ep *ep)
419 set_bit(RELEASE_RESOURCES, &ep->com.flags);
422 * If we have a hwtid, then remove it from the idr table
423 * so lookups will no longer find this endpoint. Otherwise
424 * we have a race where one thread finds the ep ptr just
425 * before the other thread is freeing the ep memory.
427 if (ep->hwtid != -1)
428 remove_ep_tid(ep);
429 c4iw_put_ep(&ep->com);
432 static int status2errno(int status)
434 switch (status) {
435 case CPL_ERR_NONE:
436 return 0;
437 case CPL_ERR_CONN_RESET:
438 return -ECONNRESET;
439 case CPL_ERR_ARP_MISS:
440 return -EHOSTUNREACH;
441 case CPL_ERR_CONN_TIMEDOUT:
442 return -ETIMEDOUT;
443 case CPL_ERR_TCAM_FULL:
444 return -ENOMEM;
445 case CPL_ERR_CONN_EXIST:
446 return -EADDRINUSE;
447 default:
448 return -EIO;
453 * Try and reuse skbs already allocated...
455 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
457 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
458 skb_trim(skb, 0);
459 skb_get(skb);
460 skb_reset_transport_header(skb);
461 } else {
462 skb = alloc_skb(len, gfp);
463 if (!skb)
464 return NULL;
466 t4_set_arp_err_handler(skb, NULL, NULL);
467 return skb;
470 static struct net_device *get_real_dev(struct net_device *egress_dev)
472 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
475 static void arp_failure_discard(void *handle, struct sk_buff *skb)
477 pr_err("ARP failure\n");
478 kfree_skb(skb);
481 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
483 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
486 enum {
487 NUM_FAKE_CPLS = 2,
488 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
489 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
492 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
494 struct c4iw_ep *ep;
496 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
497 release_ep_resources(ep);
498 return 0;
501 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
503 struct c4iw_ep *ep;
505 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
506 c4iw_put_ep(&ep->parent_ep->com);
507 release_ep_resources(ep);
508 return 0;
512 * Fake up a special CPL opcode and call sched() so process_work() will call
513 * _put_ep_safe() in a safe context to free the ep resources. This is needed
514 * because ARP error handlers are called in an ATOMIC context, and
515 * _c4iw_free_ep() needs to block.
517 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
518 int cpl)
520 struct cpl_act_establish *rpl = cplhdr(skb);
522 /* Set our special ARP_FAILURE opcode */
523 rpl->ot.opcode = cpl;
526 * Save ep in the skb->cb area, after where sched() will save the dev
527 * ptr.
529 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
530 sched(ep->com.dev, skb);
533 /* Handle an ARP failure for an accept */
534 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
536 struct c4iw_ep *ep = handle;
538 pr_err("ARP failure during accept - tid %u - dropping connection\n",
539 ep->hwtid);
541 __state_set(&ep->com, DEAD);
542 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
546 * Handle an ARP failure for an active open.
548 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
550 struct c4iw_ep *ep = handle;
552 pr_err("ARP failure during connect\n");
553 connect_reply_upcall(ep, -EHOSTUNREACH);
554 __state_set(&ep->com, DEAD);
555 if (ep->com.remote_addr.ss_family == AF_INET6) {
556 struct sockaddr_in6 *sin6 =
557 (struct sockaddr_in6 *)&ep->com.local_addr;
558 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
559 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
561 xa_erase_irq(&ep->com.dev->atids, ep->atid);
562 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
563 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
567 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
568 * and send it along.
570 static void abort_arp_failure(void *handle, struct sk_buff *skb)
572 int ret;
573 struct c4iw_ep *ep = handle;
574 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
575 struct cpl_abort_req *req = cplhdr(skb);
577 pr_debug("rdev %p\n", rdev);
578 req->cmd = CPL_ABORT_NO_RST;
579 skb_get(skb);
580 ret = c4iw_ofld_send(rdev, skb);
581 if (ret) {
582 __state_set(&ep->com, DEAD);
583 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
584 } else
585 kfree_skb(skb);
588 static int send_flowc(struct c4iw_ep *ep)
590 struct fw_flowc_wr *flowc;
591 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
592 u16 vlan = ep->l2t->vlan;
593 int nparams;
594 int flowclen, flowclen16;
596 if (WARN_ON(!skb))
597 return -ENOMEM;
599 if (vlan == CPL_L2T_VLAN_NONE)
600 nparams = 9;
601 else
602 nparams = 10;
604 flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
605 flowclen16 = DIV_ROUND_UP(flowclen, 16);
606 flowclen = flowclen16 * 16;
608 flowc = __skb_put(skb, flowclen);
609 memset(flowc, 0, flowclen);
611 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
612 FW_FLOWC_WR_NPARAMS_V(nparams));
613 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
614 FW_WR_FLOWID_V(ep->hwtid));
616 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
617 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
618 (ep->com.dev->rdev.lldi.pf));
619 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
620 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
621 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
622 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
623 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
624 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
625 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
626 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
627 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
628 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
629 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
630 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
631 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
632 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
633 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
634 flowc->mnemval[8].val = cpu_to_be32(ep->snd_wscale);
635 if (nparams == 10) {
636 u16 pri;
637 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
638 flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
639 flowc->mnemval[9].val = cpu_to_be32(pri);
642 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
643 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
646 static int send_halfclose(struct c4iw_ep *ep)
648 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
649 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
651 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
652 if (WARN_ON(!skb))
653 return -ENOMEM;
655 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
656 NULL, arp_failure_discard);
658 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
661 static void read_tcb(struct c4iw_ep *ep)
663 struct sk_buff *skb;
664 struct cpl_get_tcb *req;
665 int wrlen = roundup(sizeof(*req), 16);
667 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
668 if (WARN_ON(!skb))
669 return;
671 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
672 req = (struct cpl_get_tcb *) skb_put(skb, wrlen);
673 memset(req, 0, wrlen);
674 INIT_TP_WR(req, ep->hwtid);
675 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid));
676 req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid));
679 * keep a ref on the ep so the tcb is not unlocked before this
680 * cpl completes. The ref is released in read_tcb_rpl().
682 c4iw_get_ep(&ep->com);
683 if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb)))
684 c4iw_put_ep(&ep->com);
687 static int send_abort_req(struct c4iw_ep *ep)
689 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
690 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
692 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
693 if (WARN_ON(!req_skb))
694 return -ENOMEM;
696 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
697 ep, abort_arp_failure);
699 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
702 static int send_abort(struct c4iw_ep *ep)
704 if (!ep->com.qp || !ep->com.qp->srq) {
705 send_abort_req(ep);
706 return 0;
708 set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags);
709 read_tcb(ep);
710 return 0;
713 static int send_connect(struct c4iw_ep *ep)
715 struct cpl_act_open_req *req = NULL;
716 struct cpl_t5_act_open_req *t5req = NULL;
717 struct cpl_t6_act_open_req *t6req = NULL;
718 struct cpl_act_open_req6 *req6 = NULL;
719 struct cpl_t5_act_open_req6 *t5req6 = NULL;
720 struct cpl_t6_act_open_req6 *t6req6 = NULL;
721 struct sk_buff *skb;
722 u64 opt0;
723 u32 opt2;
724 unsigned int mtu_idx;
725 u32 wscale;
726 int win, sizev4, sizev6, wrlen;
727 struct sockaddr_in *la = (struct sockaddr_in *)
728 &ep->com.local_addr;
729 struct sockaddr_in *ra = (struct sockaddr_in *)
730 &ep->com.remote_addr;
731 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
732 &ep->com.local_addr;
733 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
734 &ep->com.remote_addr;
735 int ret;
736 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
737 u32 isn = (prandom_u32() & ~7UL) - 1;
738 struct net_device *netdev;
739 u64 params;
741 netdev = ep->com.dev->rdev.lldi.ports[0];
743 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
744 case CHELSIO_T4:
745 sizev4 = sizeof(struct cpl_act_open_req);
746 sizev6 = sizeof(struct cpl_act_open_req6);
747 break;
748 case CHELSIO_T5:
749 sizev4 = sizeof(struct cpl_t5_act_open_req);
750 sizev6 = sizeof(struct cpl_t5_act_open_req6);
751 break;
752 case CHELSIO_T6:
753 sizev4 = sizeof(struct cpl_t6_act_open_req);
754 sizev6 = sizeof(struct cpl_t6_act_open_req6);
755 break;
756 default:
757 pr_err("T%d Chip is not supported\n",
758 CHELSIO_CHIP_VERSION(adapter_type));
759 return -EINVAL;
762 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
763 roundup(sizev4, 16) :
764 roundup(sizev6, 16);
766 pr_debug("ep %p atid %u\n", ep, ep->atid);
768 skb = get_skb(NULL, wrlen, GFP_KERNEL);
769 if (!skb) {
770 pr_err("%s - failed to alloc skb\n", __func__);
771 return -ENOMEM;
773 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
775 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
776 enable_tcp_timestamps,
777 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
778 wscale = cxgb_compute_wscale(rcv_win);
781 * Specify the largest window that will fit in opt0. The
782 * remainder will be specified in the rx_data_ack.
784 win = ep->rcv_win >> 10;
785 if (win > RCV_BUFSIZ_M)
786 win = RCV_BUFSIZ_M;
788 opt0 = (nocong ? NO_CONG_F : 0) |
789 KEEP_ALIVE_F |
790 DELACK_F |
791 WND_SCALE_V(wscale) |
792 MSS_IDX_V(mtu_idx) |
793 L2T_IDX_V(ep->l2t->idx) |
794 TX_CHAN_V(ep->tx_chan) |
795 SMAC_SEL_V(ep->smac_idx) |
796 DSCP_V(ep->tos >> 2) |
797 ULP_MODE_V(ULP_MODE_TCPDDP) |
798 RCV_BUFSIZ_V(win);
799 opt2 = RX_CHANNEL_V(0) |
800 CCTRL_ECN_V(enable_ecn) |
801 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
802 if (enable_tcp_timestamps)
803 opt2 |= TSTAMPS_EN_F;
804 if (enable_tcp_sack)
805 opt2 |= SACK_EN_F;
806 if (wscale && enable_tcp_window_scaling)
807 opt2 |= WND_SCALE_EN_F;
808 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
809 if (peer2peer)
810 isn += 4;
812 opt2 |= T5_OPT_2_VALID_F;
813 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
814 opt2 |= T5_ISS_F;
817 params = cxgb4_select_ntuple(netdev, ep->l2t);
819 if (ep->com.remote_addr.ss_family == AF_INET6)
820 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
821 (const u32 *)&la6->sin6_addr.s6_addr, 1);
823 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
825 if (ep->com.remote_addr.ss_family == AF_INET) {
826 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
827 case CHELSIO_T4:
828 req = skb_put(skb, wrlen);
829 INIT_TP_WR(req, 0);
830 break;
831 case CHELSIO_T5:
832 t5req = skb_put(skb, wrlen);
833 INIT_TP_WR(t5req, 0);
834 req = (struct cpl_act_open_req *)t5req;
835 break;
836 case CHELSIO_T6:
837 t6req = skb_put(skb, wrlen);
838 INIT_TP_WR(t6req, 0);
839 req = (struct cpl_act_open_req *)t6req;
840 t5req = (struct cpl_t5_act_open_req *)t6req;
841 break;
842 default:
843 pr_err("T%d Chip is not supported\n",
844 CHELSIO_CHIP_VERSION(adapter_type));
845 ret = -EINVAL;
846 goto clip_release;
849 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
850 ((ep->rss_qid<<14) | ep->atid)));
851 req->local_port = la->sin_port;
852 req->peer_port = ra->sin_port;
853 req->local_ip = la->sin_addr.s_addr;
854 req->peer_ip = ra->sin_addr.s_addr;
855 req->opt0 = cpu_to_be64(opt0);
857 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
858 req->params = cpu_to_be32(params);
859 req->opt2 = cpu_to_be32(opt2);
860 } else {
861 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
862 t5req->params =
863 cpu_to_be64(FILTER_TUPLE_V(params));
864 t5req->rsvd = cpu_to_be32(isn);
865 pr_debug("snd_isn %u\n", t5req->rsvd);
866 t5req->opt2 = cpu_to_be32(opt2);
867 } else {
868 t6req->params =
869 cpu_to_be64(FILTER_TUPLE_V(params));
870 t6req->rsvd = cpu_to_be32(isn);
871 pr_debug("snd_isn %u\n", t6req->rsvd);
872 t6req->opt2 = cpu_to_be32(opt2);
875 } else {
876 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
877 case CHELSIO_T4:
878 req6 = skb_put(skb, wrlen);
879 INIT_TP_WR(req6, 0);
880 break;
881 case CHELSIO_T5:
882 t5req6 = skb_put(skb, wrlen);
883 INIT_TP_WR(t5req6, 0);
884 req6 = (struct cpl_act_open_req6 *)t5req6;
885 break;
886 case CHELSIO_T6:
887 t6req6 = skb_put(skb, wrlen);
888 INIT_TP_WR(t6req6, 0);
889 req6 = (struct cpl_act_open_req6 *)t6req6;
890 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
891 break;
892 default:
893 pr_err("T%d Chip is not supported\n",
894 CHELSIO_CHIP_VERSION(adapter_type));
895 ret = -EINVAL;
896 goto clip_release;
899 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
900 ((ep->rss_qid<<14)|ep->atid)));
901 req6->local_port = la6->sin6_port;
902 req6->peer_port = ra6->sin6_port;
903 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
904 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
905 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
906 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
907 req6->opt0 = cpu_to_be64(opt0);
909 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
910 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
911 ep->l2t));
912 req6->opt2 = cpu_to_be32(opt2);
913 } else {
914 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
915 t5req6->params =
916 cpu_to_be64(FILTER_TUPLE_V(params));
917 t5req6->rsvd = cpu_to_be32(isn);
918 pr_debug("snd_isn %u\n", t5req6->rsvd);
919 t5req6->opt2 = cpu_to_be32(opt2);
920 } else {
921 t6req6->params =
922 cpu_to_be64(FILTER_TUPLE_V(params));
923 t6req6->rsvd = cpu_to_be32(isn);
924 pr_debug("snd_isn %u\n", t6req6->rsvd);
925 t6req6->opt2 = cpu_to_be32(opt2);
931 set_bit(ACT_OPEN_REQ, &ep->com.history);
932 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
933 clip_release:
934 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
935 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
936 (const u32 *)&la6->sin6_addr.s6_addr, 1);
937 return ret;
940 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
941 u8 mpa_rev_to_use)
943 int mpalen, wrlen, ret;
944 struct fw_ofld_tx_data_wr *req;
945 struct mpa_message *mpa;
946 struct mpa_v2_conn_params mpa_v2_params;
948 pr_debug("ep %p tid %u pd_len %d\n",
949 ep, ep->hwtid, ep->plen);
951 mpalen = sizeof(*mpa) + ep->plen;
952 if (mpa_rev_to_use == 2)
953 mpalen += sizeof(struct mpa_v2_conn_params);
954 wrlen = roundup(mpalen + sizeof(*req), 16);
955 skb = get_skb(skb, wrlen, GFP_KERNEL);
956 if (!skb) {
957 connect_reply_upcall(ep, -ENOMEM);
958 return -ENOMEM;
960 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
962 req = skb_put_zero(skb, wrlen);
963 req->op_to_immdlen = cpu_to_be32(
964 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
965 FW_WR_COMPL_F |
966 FW_WR_IMMDLEN_V(mpalen));
967 req->flowid_len16 = cpu_to_be32(
968 FW_WR_FLOWID_V(ep->hwtid) |
969 FW_WR_LEN16_V(wrlen >> 4));
970 req->plen = cpu_to_be32(mpalen);
971 req->tunnel_to_proxy = cpu_to_be32(
972 FW_OFLD_TX_DATA_WR_FLUSH_F |
973 FW_OFLD_TX_DATA_WR_SHOVE_F);
975 mpa = (struct mpa_message *)(req + 1);
976 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
978 mpa->flags = 0;
979 if (crc_enabled)
980 mpa->flags |= MPA_CRC;
981 if (markers_enabled) {
982 mpa->flags |= MPA_MARKERS;
983 ep->mpa_attr.recv_marker_enabled = 1;
984 } else {
985 ep->mpa_attr.recv_marker_enabled = 0;
987 if (mpa_rev_to_use == 2)
988 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
990 mpa->private_data_size = htons(ep->plen);
991 mpa->revision = mpa_rev_to_use;
992 if (mpa_rev_to_use == 1) {
993 ep->tried_with_mpa_v1 = 1;
994 ep->retry_with_mpa_v1 = 0;
997 if (mpa_rev_to_use == 2) {
998 mpa->private_data_size =
999 htons(ntohs(mpa->private_data_size) +
1000 sizeof(struct mpa_v2_conn_params));
1001 pr_debug("initiator ird %u ord %u\n", ep->ird,
1002 ep->ord);
1003 mpa_v2_params.ird = htons((u16)ep->ird);
1004 mpa_v2_params.ord = htons((u16)ep->ord);
1006 if (peer2peer) {
1007 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1008 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1009 mpa_v2_params.ord |=
1010 htons(MPA_V2_RDMA_WRITE_RTR);
1011 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1012 mpa_v2_params.ord |=
1013 htons(MPA_V2_RDMA_READ_RTR);
1015 memcpy(mpa->private_data, &mpa_v2_params,
1016 sizeof(struct mpa_v2_conn_params));
1018 if (ep->plen)
1019 memcpy(mpa->private_data +
1020 sizeof(struct mpa_v2_conn_params),
1021 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1022 } else
1023 if (ep->plen)
1024 memcpy(mpa->private_data,
1025 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1028 * Reference the mpa skb. This ensures the data area
1029 * will remain in memory until the hw acks the tx.
1030 * Function fw4_ack() will deref it.
1032 skb_get(skb);
1033 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1034 ep->mpa_skb = skb;
1035 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1036 if (ret)
1037 return ret;
1038 start_ep_timer(ep);
1039 __state_set(&ep->com, MPA_REQ_SENT);
1040 ep->mpa_attr.initiator = 1;
1041 ep->snd_seq += mpalen;
1042 return ret;
1045 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1047 int mpalen, wrlen;
1048 struct fw_ofld_tx_data_wr *req;
1049 struct mpa_message *mpa;
1050 struct sk_buff *skb;
1051 struct mpa_v2_conn_params mpa_v2_params;
1053 pr_debug("ep %p tid %u pd_len %d\n",
1054 ep, ep->hwtid, ep->plen);
1056 mpalen = sizeof(*mpa) + plen;
1057 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1058 mpalen += sizeof(struct mpa_v2_conn_params);
1059 wrlen = roundup(mpalen + sizeof(*req), 16);
1061 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1062 if (!skb) {
1063 pr_err("%s - cannot alloc skb!\n", __func__);
1064 return -ENOMEM;
1066 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1068 req = skb_put_zero(skb, wrlen);
1069 req->op_to_immdlen = cpu_to_be32(
1070 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1071 FW_WR_COMPL_F |
1072 FW_WR_IMMDLEN_V(mpalen));
1073 req->flowid_len16 = cpu_to_be32(
1074 FW_WR_FLOWID_V(ep->hwtid) |
1075 FW_WR_LEN16_V(wrlen >> 4));
1076 req->plen = cpu_to_be32(mpalen);
1077 req->tunnel_to_proxy = cpu_to_be32(
1078 FW_OFLD_TX_DATA_WR_FLUSH_F |
1079 FW_OFLD_TX_DATA_WR_SHOVE_F);
1081 mpa = (struct mpa_message *)(req + 1);
1082 memset(mpa, 0, sizeof(*mpa));
1083 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1084 mpa->flags = MPA_REJECT;
1085 mpa->revision = ep->mpa_attr.version;
1086 mpa->private_data_size = htons(plen);
1088 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1089 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1090 mpa->private_data_size =
1091 htons(ntohs(mpa->private_data_size) +
1092 sizeof(struct mpa_v2_conn_params));
1093 mpa_v2_params.ird = htons(((u16)ep->ird) |
1094 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1095 0));
1096 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1097 (p2p_type ==
1098 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1099 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1100 FW_RI_INIT_P2PTYPE_READ_REQ ?
1101 MPA_V2_RDMA_READ_RTR : 0) : 0));
1102 memcpy(mpa->private_data, &mpa_v2_params,
1103 sizeof(struct mpa_v2_conn_params));
1105 if (ep->plen)
1106 memcpy(mpa->private_data +
1107 sizeof(struct mpa_v2_conn_params), pdata, plen);
1108 } else
1109 if (plen)
1110 memcpy(mpa->private_data, pdata, plen);
1113 * Reference the mpa skb again. This ensures the data area
1114 * will remain in memory until the hw acks the tx.
1115 * Function fw4_ack() will deref it.
1117 skb_get(skb);
1118 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1119 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1120 ep->mpa_skb = skb;
1121 ep->snd_seq += mpalen;
1122 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1125 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1127 int mpalen, wrlen;
1128 struct fw_ofld_tx_data_wr *req;
1129 struct mpa_message *mpa;
1130 struct sk_buff *skb;
1131 struct mpa_v2_conn_params mpa_v2_params;
1133 pr_debug("ep %p tid %u pd_len %d\n",
1134 ep, ep->hwtid, ep->plen);
1136 mpalen = sizeof(*mpa) + plen;
1137 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1138 mpalen += sizeof(struct mpa_v2_conn_params);
1139 wrlen = roundup(mpalen + sizeof(*req), 16);
1141 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1142 if (!skb) {
1143 pr_err("%s - cannot alloc skb!\n", __func__);
1144 return -ENOMEM;
1146 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1148 req = skb_put_zero(skb, wrlen);
1149 req->op_to_immdlen = cpu_to_be32(
1150 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1151 FW_WR_COMPL_F |
1152 FW_WR_IMMDLEN_V(mpalen));
1153 req->flowid_len16 = cpu_to_be32(
1154 FW_WR_FLOWID_V(ep->hwtid) |
1155 FW_WR_LEN16_V(wrlen >> 4));
1156 req->plen = cpu_to_be32(mpalen);
1157 req->tunnel_to_proxy = cpu_to_be32(
1158 FW_OFLD_TX_DATA_WR_FLUSH_F |
1159 FW_OFLD_TX_DATA_WR_SHOVE_F);
1161 mpa = (struct mpa_message *)(req + 1);
1162 memset(mpa, 0, sizeof(*mpa));
1163 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1164 mpa->flags = 0;
1165 if (ep->mpa_attr.crc_enabled)
1166 mpa->flags |= MPA_CRC;
1167 if (ep->mpa_attr.recv_marker_enabled)
1168 mpa->flags |= MPA_MARKERS;
1169 mpa->revision = ep->mpa_attr.version;
1170 mpa->private_data_size = htons(plen);
1172 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1173 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1174 mpa->private_data_size =
1175 htons(ntohs(mpa->private_data_size) +
1176 sizeof(struct mpa_v2_conn_params));
1177 mpa_v2_params.ird = htons((u16)ep->ird);
1178 mpa_v2_params.ord = htons((u16)ep->ord);
1179 if (peer2peer && (ep->mpa_attr.p2p_type !=
1180 FW_RI_INIT_P2PTYPE_DISABLED)) {
1181 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1183 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1184 mpa_v2_params.ord |=
1185 htons(MPA_V2_RDMA_WRITE_RTR);
1186 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1187 mpa_v2_params.ord |=
1188 htons(MPA_V2_RDMA_READ_RTR);
1191 memcpy(mpa->private_data, &mpa_v2_params,
1192 sizeof(struct mpa_v2_conn_params));
1194 if (ep->plen)
1195 memcpy(mpa->private_data +
1196 sizeof(struct mpa_v2_conn_params), pdata, plen);
1197 } else
1198 if (plen)
1199 memcpy(mpa->private_data, pdata, plen);
1202 * Reference the mpa skb. This ensures the data area
1203 * will remain in memory until the hw acks the tx.
1204 * Function fw4_ack() will deref it.
1206 skb_get(skb);
1207 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1208 ep->mpa_skb = skb;
1209 __state_set(&ep->com, MPA_REP_SENT);
1210 ep->snd_seq += mpalen;
1211 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1214 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1216 struct c4iw_ep *ep;
1217 struct cpl_act_establish *req = cplhdr(skb);
1218 unsigned short tcp_opt = ntohs(req->tcp_opt);
1219 unsigned int tid = GET_TID(req);
1220 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1221 struct tid_info *t = dev->rdev.lldi.tids;
1222 int ret;
1224 ep = lookup_atid(t, atid);
1226 pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid,
1227 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1229 mutex_lock(&ep->com.mutex);
1230 dst_confirm(ep->dst);
1232 /* setup the hwtid for this connection */
1233 ep->hwtid = tid;
1234 cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1235 insert_ep_tid(ep);
1237 ep->snd_seq = be32_to_cpu(req->snd_isn);
1238 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1239 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1241 set_emss(ep, tcp_opt);
1243 /* dealloc the atid */
1244 xa_erase_irq(&ep->com.dev->atids, atid);
1245 cxgb4_free_atid(t, atid);
1246 set_bit(ACT_ESTAB, &ep->com.history);
1248 /* start MPA negotiation */
1249 ret = send_flowc(ep);
1250 if (ret)
1251 goto err;
1252 if (ep->retry_with_mpa_v1)
1253 ret = send_mpa_req(ep, skb, 1);
1254 else
1255 ret = send_mpa_req(ep, skb, mpa_rev);
1256 if (ret)
1257 goto err;
1258 mutex_unlock(&ep->com.mutex);
1259 return 0;
1260 err:
1261 mutex_unlock(&ep->com.mutex);
1262 connect_reply_upcall(ep, -ENOMEM);
1263 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1264 return 0;
1267 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1269 struct iw_cm_event event;
1271 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1272 memset(&event, 0, sizeof(event));
1273 event.event = IW_CM_EVENT_CLOSE;
1274 event.status = status;
1275 if (ep->com.cm_id) {
1276 pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1277 ep, ep->com.cm_id, ep->hwtid);
1278 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1279 deref_cm_id(&ep->com);
1280 set_bit(CLOSE_UPCALL, &ep->com.history);
1284 static void peer_close_upcall(struct c4iw_ep *ep)
1286 struct iw_cm_event event;
1288 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1289 memset(&event, 0, sizeof(event));
1290 event.event = IW_CM_EVENT_DISCONNECT;
1291 if (ep->com.cm_id) {
1292 pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1293 ep, ep->com.cm_id, ep->hwtid);
1294 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1295 set_bit(DISCONN_UPCALL, &ep->com.history);
1299 static void peer_abort_upcall(struct c4iw_ep *ep)
1301 struct iw_cm_event event;
1303 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1304 memset(&event, 0, sizeof(event));
1305 event.event = IW_CM_EVENT_CLOSE;
1306 event.status = -ECONNRESET;
1307 if (ep->com.cm_id) {
1308 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1309 ep->com.cm_id, ep->hwtid);
1310 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1311 deref_cm_id(&ep->com);
1312 set_bit(ABORT_UPCALL, &ep->com.history);
1316 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1318 struct iw_cm_event event;
1320 pr_debug("ep %p tid %u status %d\n",
1321 ep, ep->hwtid, status);
1322 memset(&event, 0, sizeof(event));
1323 event.event = IW_CM_EVENT_CONNECT_REPLY;
1324 event.status = status;
1325 memcpy(&event.local_addr, &ep->com.local_addr,
1326 sizeof(ep->com.local_addr));
1327 memcpy(&event.remote_addr, &ep->com.remote_addr,
1328 sizeof(ep->com.remote_addr));
1330 if ((status == 0) || (status == -ECONNREFUSED)) {
1331 if (!ep->tried_with_mpa_v1) {
1332 /* this means MPA_v2 is used */
1333 event.ord = ep->ird;
1334 event.ird = ep->ord;
1335 event.private_data_len = ep->plen -
1336 sizeof(struct mpa_v2_conn_params);
1337 event.private_data = ep->mpa_pkt +
1338 sizeof(struct mpa_message) +
1339 sizeof(struct mpa_v2_conn_params);
1340 } else {
1341 /* this means MPA_v1 is used */
1342 event.ord = cur_max_read_depth(ep->com.dev);
1343 event.ird = cur_max_read_depth(ep->com.dev);
1344 event.private_data_len = ep->plen;
1345 event.private_data = ep->mpa_pkt +
1346 sizeof(struct mpa_message);
1350 pr_debug("ep %p tid %u status %d\n", ep,
1351 ep->hwtid, status);
1352 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1353 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1355 if (status < 0)
1356 deref_cm_id(&ep->com);
1359 static int connect_request_upcall(struct c4iw_ep *ep)
1361 struct iw_cm_event event;
1362 int ret;
1364 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1365 memset(&event, 0, sizeof(event));
1366 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1367 memcpy(&event.local_addr, &ep->com.local_addr,
1368 sizeof(ep->com.local_addr));
1369 memcpy(&event.remote_addr, &ep->com.remote_addr,
1370 sizeof(ep->com.remote_addr));
1371 event.provider_data = ep;
1372 if (!ep->tried_with_mpa_v1) {
1373 /* this means MPA_v2 is used */
1374 event.ord = ep->ord;
1375 event.ird = ep->ird;
1376 event.private_data_len = ep->plen -
1377 sizeof(struct mpa_v2_conn_params);
1378 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1379 sizeof(struct mpa_v2_conn_params);
1380 } else {
1381 /* this means MPA_v1 is used. Send max supported */
1382 event.ord = cur_max_read_depth(ep->com.dev);
1383 event.ird = cur_max_read_depth(ep->com.dev);
1384 event.private_data_len = ep->plen;
1385 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1387 c4iw_get_ep(&ep->com);
1388 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1389 &event);
1390 if (ret)
1391 c4iw_put_ep(&ep->com);
1392 set_bit(CONNREQ_UPCALL, &ep->com.history);
1393 c4iw_put_ep(&ep->parent_ep->com);
1394 return ret;
1397 static void established_upcall(struct c4iw_ep *ep)
1399 struct iw_cm_event event;
1401 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1402 memset(&event, 0, sizeof(event));
1403 event.event = IW_CM_EVENT_ESTABLISHED;
1404 event.ird = ep->ord;
1405 event.ord = ep->ird;
1406 if (ep->com.cm_id) {
1407 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1408 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1409 set_bit(ESTAB_UPCALL, &ep->com.history);
1413 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1415 struct sk_buff *skb;
1416 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1417 u32 credit_dack;
1419 pr_debug("ep %p tid %u credits %u\n",
1420 ep, ep->hwtid, credits);
1421 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1422 if (!skb) {
1423 pr_err("update_rx_credits - cannot alloc skb!\n");
1424 return 0;
1428 * If we couldn't specify the entire rcv window at connection setup
1429 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1430 * then add the overage in to the credits returned.
1432 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1433 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1435 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1436 RX_DACK_MODE_V(dack_mode);
1438 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1439 credit_dack);
1441 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1442 return credits;
1445 #define RELAXED_IRD_NEGOTIATION 1
1448 * process_mpa_reply - process streaming mode MPA reply
1450 * Returns:
1452 * 0 upon success indicating a connect request was delivered to the ULP
1453 * or the mpa request is incomplete but valid so far.
1455 * 1 if a failure requires the caller to close the connection.
1457 * 2 if a failure requires the caller to abort the connection.
1459 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1461 struct mpa_message *mpa;
1462 struct mpa_v2_conn_params *mpa_v2_params;
1463 u16 plen;
1464 u16 resp_ird, resp_ord;
1465 u8 rtr_mismatch = 0, insuff_ird = 0;
1466 struct c4iw_qp_attributes attrs;
1467 enum c4iw_qp_attr_mask mask;
1468 int err;
1469 int disconnect = 0;
1471 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1474 * If we get more than the supported amount of private data
1475 * then we must fail this connection.
1477 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1478 err = -EINVAL;
1479 goto err_stop_timer;
1483 * copy the new data into our accumulation buffer.
1485 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1486 skb->len);
1487 ep->mpa_pkt_len += skb->len;
1490 * if we don't even have the mpa message, then bail.
1492 if (ep->mpa_pkt_len < sizeof(*mpa))
1493 return 0;
1494 mpa = (struct mpa_message *) ep->mpa_pkt;
1496 /* Validate MPA header. */
1497 if (mpa->revision > mpa_rev) {
1498 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1499 __func__, mpa_rev, mpa->revision);
1500 err = -EPROTO;
1501 goto err_stop_timer;
1503 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1504 err = -EPROTO;
1505 goto err_stop_timer;
1508 plen = ntohs(mpa->private_data_size);
1511 * Fail if there's too much private data.
1513 if (plen > MPA_MAX_PRIVATE_DATA) {
1514 err = -EPROTO;
1515 goto err_stop_timer;
1519 * If plen does not account for pkt size
1521 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1522 err = -EPROTO;
1523 goto err_stop_timer;
1526 ep->plen = (u8) plen;
1529 * If we don't have all the pdata yet, then bail.
1530 * We'll continue process when more data arrives.
1532 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1533 return 0;
1535 if (mpa->flags & MPA_REJECT) {
1536 err = -ECONNREFUSED;
1537 goto err_stop_timer;
1541 * Stop mpa timer. If it expired, then
1542 * we ignore the MPA reply. process_timeout()
1543 * will abort the connection.
1545 if (stop_ep_timer(ep))
1546 return 0;
1549 * If we get here we have accumulated the entire mpa
1550 * start reply message including private data. And
1551 * the MPA header is valid.
1553 __state_set(&ep->com, FPDU_MODE);
1554 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1555 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1556 ep->mpa_attr.version = mpa->revision;
1557 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1559 if (mpa->revision == 2) {
1560 ep->mpa_attr.enhanced_rdma_conn =
1561 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1562 if (ep->mpa_attr.enhanced_rdma_conn) {
1563 mpa_v2_params = (struct mpa_v2_conn_params *)
1564 (ep->mpa_pkt + sizeof(*mpa));
1565 resp_ird = ntohs(mpa_v2_params->ird) &
1566 MPA_V2_IRD_ORD_MASK;
1567 resp_ord = ntohs(mpa_v2_params->ord) &
1568 MPA_V2_IRD_ORD_MASK;
1569 pr_debug("responder ird %u ord %u ep ird %u ord %u\n",
1570 resp_ird, resp_ord, ep->ird, ep->ord);
1573 * This is a double-check. Ideally, below checks are
1574 * not required since ird/ord stuff has been taken
1575 * care of in c4iw_accept_cr
1577 if (ep->ird < resp_ord) {
1578 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1579 ep->com.dev->rdev.lldi.max_ordird_qp)
1580 ep->ird = resp_ord;
1581 else
1582 insuff_ird = 1;
1583 } else if (ep->ird > resp_ord) {
1584 ep->ird = resp_ord;
1586 if (ep->ord > resp_ird) {
1587 if (RELAXED_IRD_NEGOTIATION)
1588 ep->ord = resp_ird;
1589 else
1590 insuff_ird = 1;
1592 if (insuff_ird) {
1593 err = -ENOMEM;
1594 ep->ird = resp_ord;
1595 ep->ord = resp_ird;
1598 if (ntohs(mpa_v2_params->ird) &
1599 MPA_V2_PEER2PEER_MODEL) {
1600 if (ntohs(mpa_v2_params->ord) &
1601 MPA_V2_RDMA_WRITE_RTR)
1602 ep->mpa_attr.p2p_type =
1603 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1604 else if (ntohs(mpa_v2_params->ord) &
1605 MPA_V2_RDMA_READ_RTR)
1606 ep->mpa_attr.p2p_type =
1607 FW_RI_INIT_P2PTYPE_READ_REQ;
1610 } else if (mpa->revision == 1)
1611 if (peer2peer)
1612 ep->mpa_attr.p2p_type = p2p_type;
1614 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1615 ep->mpa_attr.crc_enabled,
1616 ep->mpa_attr.recv_marker_enabled,
1617 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1618 ep->mpa_attr.p2p_type, p2p_type);
1621 * If responder's RTR does not match with that of initiator, assign
1622 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1623 * generated when moving QP to RTS state.
1624 * A TERM message will be sent after QP has moved to RTS state
1626 if ((ep->mpa_attr.version == 2) && peer2peer &&
1627 (ep->mpa_attr.p2p_type != p2p_type)) {
1628 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1629 rtr_mismatch = 1;
1632 attrs.mpa_attr = ep->mpa_attr;
1633 attrs.max_ird = ep->ird;
1634 attrs.max_ord = ep->ord;
1635 attrs.llp_stream_handle = ep;
1636 attrs.next_state = C4IW_QP_STATE_RTS;
1638 mask = C4IW_QP_ATTR_NEXT_STATE |
1639 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1640 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1642 /* bind QP and TID with INIT_WR */
1643 err = c4iw_modify_qp(ep->com.qp->rhp,
1644 ep->com.qp, mask, &attrs, 1);
1645 if (err)
1646 goto err;
1649 * If responder's RTR requirement did not match with what initiator
1650 * supports, generate TERM message
1652 if (rtr_mismatch) {
1653 pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1654 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1655 attrs.ecode = MPA_NOMATCH_RTR;
1656 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1657 attrs.send_term = 1;
1658 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1659 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1660 err = -ENOMEM;
1661 disconnect = 1;
1662 goto out;
1666 * Generate TERM if initiator IRD is not sufficient for responder
1667 * provided ORD. Currently, we do the same behaviour even when
1668 * responder provided IRD is also not sufficient as regards to
1669 * initiator ORD.
1671 if (insuff_ird) {
1672 pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1673 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1674 attrs.ecode = MPA_INSUFF_IRD;
1675 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1676 attrs.send_term = 1;
1677 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1678 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1679 err = -ENOMEM;
1680 disconnect = 1;
1681 goto out;
1683 goto out;
1684 err_stop_timer:
1685 stop_ep_timer(ep);
1686 err:
1687 disconnect = 2;
1688 out:
1689 connect_reply_upcall(ep, err);
1690 return disconnect;
1694 * process_mpa_request - process streaming mode MPA request
1696 * Returns:
1698 * 0 upon success indicating a connect request was delivered to the ULP
1699 * or the mpa request is incomplete but valid so far.
1701 * 1 if a failure requires the caller to close the connection.
1703 * 2 if a failure requires the caller to abort the connection.
1705 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1707 struct mpa_message *mpa;
1708 struct mpa_v2_conn_params *mpa_v2_params;
1709 u16 plen;
1711 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1714 * If we get more than the supported amount of private data
1715 * then we must fail this connection.
1717 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1718 goto err_stop_timer;
1720 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1723 * Copy the new data into our accumulation buffer.
1725 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1726 skb->len);
1727 ep->mpa_pkt_len += skb->len;
1730 * If we don't even have the mpa message, then bail.
1731 * We'll continue process when more data arrives.
1733 if (ep->mpa_pkt_len < sizeof(*mpa))
1734 return 0;
1736 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1737 mpa = (struct mpa_message *) ep->mpa_pkt;
1740 * Validate MPA Header.
1742 if (mpa->revision > mpa_rev) {
1743 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1744 __func__, mpa_rev, mpa->revision);
1745 goto err_stop_timer;
1748 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1749 goto err_stop_timer;
1751 plen = ntohs(mpa->private_data_size);
1754 * Fail if there's too much private data.
1756 if (plen > MPA_MAX_PRIVATE_DATA)
1757 goto err_stop_timer;
1760 * If plen does not account for pkt size
1762 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1763 goto err_stop_timer;
1764 ep->plen = (u8) plen;
1767 * If we don't have all the pdata yet, then bail.
1769 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1770 return 0;
1773 * If we get here we have accumulated the entire mpa
1774 * start reply message including private data.
1776 ep->mpa_attr.initiator = 0;
1777 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1778 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1779 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1780 ep->mpa_attr.version = mpa->revision;
1781 if (mpa->revision == 1)
1782 ep->tried_with_mpa_v1 = 1;
1783 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1785 if (mpa->revision == 2) {
1786 ep->mpa_attr.enhanced_rdma_conn =
1787 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1788 if (ep->mpa_attr.enhanced_rdma_conn) {
1789 mpa_v2_params = (struct mpa_v2_conn_params *)
1790 (ep->mpa_pkt + sizeof(*mpa));
1791 ep->ird = ntohs(mpa_v2_params->ird) &
1792 MPA_V2_IRD_ORD_MASK;
1793 ep->ird = min_t(u32, ep->ird,
1794 cur_max_read_depth(ep->com.dev));
1795 ep->ord = ntohs(mpa_v2_params->ord) &
1796 MPA_V2_IRD_ORD_MASK;
1797 ep->ord = min_t(u32, ep->ord,
1798 cur_max_read_depth(ep->com.dev));
1799 pr_debug("initiator ird %u ord %u\n",
1800 ep->ird, ep->ord);
1801 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1802 if (peer2peer) {
1803 if (ntohs(mpa_v2_params->ord) &
1804 MPA_V2_RDMA_WRITE_RTR)
1805 ep->mpa_attr.p2p_type =
1806 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1807 else if (ntohs(mpa_v2_params->ord) &
1808 MPA_V2_RDMA_READ_RTR)
1809 ep->mpa_attr.p2p_type =
1810 FW_RI_INIT_P2PTYPE_READ_REQ;
1813 } else if (mpa->revision == 1)
1814 if (peer2peer)
1815 ep->mpa_attr.p2p_type = p2p_type;
1817 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1818 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1819 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1820 ep->mpa_attr.p2p_type);
1822 __state_set(&ep->com, MPA_REQ_RCVD);
1824 /* drive upcall */
1825 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1826 if (ep->parent_ep->com.state != DEAD) {
1827 if (connect_request_upcall(ep))
1828 goto err_unlock_parent;
1829 } else {
1830 goto err_unlock_parent;
1832 mutex_unlock(&ep->parent_ep->com.mutex);
1833 return 0;
1835 err_unlock_parent:
1836 mutex_unlock(&ep->parent_ep->com.mutex);
1837 goto err_out;
1838 err_stop_timer:
1839 (void)stop_ep_timer(ep);
1840 err_out:
1841 return 2;
1844 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1846 struct c4iw_ep *ep;
1847 struct cpl_rx_data *hdr = cplhdr(skb);
1848 unsigned int dlen = ntohs(hdr->len);
1849 unsigned int tid = GET_TID(hdr);
1850 __u8 status = hdr->status;
1851 int disconnect = 0;
1853 ep = get_ep_from_tid(dev, tid);
1854 if (!ep)
1855 return 0;
1856 pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen);
1857 skb_pull(skb, sizeof(*hdr));
1858 skb_trim(skb, dlen);
1859 mutex_lock(&ep->com.mutex);
1861 switch (ep->com.state) {
1862 case MPA_REQ_SENT:
1863 update_rx_credits(ep, dlen);
1864 ep->rcv_seq += dlen;
1865 disconnect = process_mpa_reply(ep, skb);
1866 break;
1867 case MPA_REQ_WAIT:
1868 update_rx_credits(ep, dlen);
1869 ep->rcv_seq += dlen;
1870 disconnect = process_mpa_request(ep, skb);
1871 break;
1872 case FPDU_MODE: {
1873 struct c4iw_qp_attributes attrs;
1875 update_rx_credits(ep, dlen);
1876 if (status)
1877 pr_err("%s Unexpected streaming data." \
1878 " qpid %u ep %p state %d tid %u status %d\n",
1879 __func__, ep->com.qp->wq.sq.qid, ep,
1880 ep->com.state, ep->hwtid, status);
1881 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1882 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1883 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1884 disconnect = 1;
1885 break;
1887 default:
1888 break;
1890 mutex_unlock(&ep->com.mutex);
1891 if (disconnect)
1892 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1893 c4iw_put_ep(&ep->com);
1894 return 0;
1897 static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx)
1899 enum chip_type adapter_type;
1901 adapter_type = ep->com.dev->rdev.lldi.adapter_type;
1904 * If this TCB had a srq buffer cached, then we must complete
1905 * it. For user mode, that means saving the srqidx in the
1906 * user/kernel status page for this qp. For kernel mode, just
1907 * synthesize the CQE now.
1909 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) {
1910 if (ep->com.qp->ibqp.uobject)
1911 t4_set_wq_in_error(&ep->com.qp->wq, srqidx);
1912 else
1913 c4iw_flush_srqidx(ep->com.qp, srqidx);
1917 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1919 u32 srqidx;
1920 struct c4iw_ep *ep;
1921 struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
1922 int release = 0;
1923 unsigned int tid = GET_TID(rpl);
1925 ep = get_ep_from_tid(dev, tid);
1926 if (!ep) {
1927 pr_warn("Abort rpl to freed endpoint\n");
1928 return 0;
1931 if (ep->com.qp && ep->com.qp->srq) {
1932 srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status));
1933 complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx);
1936 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1937 mutex_lock(&ep->com.mutex);
1938 switch (ep->com.state) {
1939 case ABORTING:
1940 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
1941 __state_set(&ep->com, DEAD);
1942 release = 1;
1943 break;
1944 default:
1945 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1946 break;
1948 mutex_unlock(&ep->com.mutex);
1950 if (release) {
1951 close_complete_upcall(ep, -ECONNRESET);
1952 release_ep_resources(ep);
1954 c4iw_put_ep(&ep->com);
1955 return 0;
1958 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1960 struct sk_buff *skb;
1961 struct fw_ofld_connection_wr *req;
1962 unsigned int mtu_idx;
1963 u32 wscale;
1964 struct sockaddr_in *sin;
1965 int win;
1967 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1968 req = __skb_put_zero(skb, sizeof(*req));
1969 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1970 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1971 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1972 ep->com.dev->rdev.lldi.ports[0],
1973 ep->l2t));
1974 sin = (struct sockaddr_in *)&ep->com.local_addr;
1975 req->le.lport = sin->sin_port;
1976 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1977 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1978 req->le.pport = sin->sin_port;
1979 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1980 req->tcb.t_state_to_astid =
1981 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1982 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1983 req->tcb.cplrxdataack_cplpassacceptrpl =
1984 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1985 req->tcb.tx_max = (__force __be32) jiffies;
1986 req->tcb.rcv_adv = htons(1);
1987 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1988 enable_tcp_timestamps,
1989 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1990 wscale = cxgb_compute_wscale(rcv_win);
1993 * Specify the largest window that will fit in opt0. The
1994 * remainder will be specified in the rx_data_ack.
1996 win = ep->rcv_win >> 10;
1997 if (win > RCV_BUFSIZ_M)
1998 win = RCV_BUFSIZ_M;
2000 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2001 (nocong ? NO_CONG_F : 0) |
2002 KEEP_ALIVE_F |
2003 DELACK_F |
2004 WND_SCALE_V(wscale) |
2005 MSS_IDX_V(mtu_idx) |
2006 L2T_IDX_V(ep->l2t->idx) |
2007 TX_CHAN_V(ep->tx_chan) |
2008 SMAC_SEL_V(ep->smac_idx) |
2009 DSCP_V(ep->tos >> 2) |
2010 ULP_MODE_V(ULP_MODE_TCPDDP) |
2011 RCV_BUFSIZ_V(win));
2012 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2013 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2014 RX_CHANNEL_V(0) |
2015 CCTRL_ECN_V(enable_ecn) |
2016 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2017 if (enable_tcp_timestamps)
2018 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2019 if (enable_tcp_sack)
2020 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2021 if (wscale && enable_tcp_window_scaling)
2022 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2023 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2024 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2025 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2026 set_bit(ACT_OFLD_CONN, &ep->com.history);
2027 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2031 * Some of the error codes above implicitly indicate that there is no TID
2032 * allocated with the result of an ACT_OPEN. We use this predicate to make
2033 * that explicit.
2035 static inline int act_open_has_tid(int status)
2037 return (status != CPL_ERR_TCAM_PARITY &&
2038 status != CPL_ERR_TCAM_MISS &&
2039 status != CPL_ERR_TCAM_FULL &&
2040 status != CPL_ERR_CONN_EXIST_SYNRECV &&
2041 status != CPL_ERR_CONN_EXIST);
2044 static char *neg_adv_str(unsigned int status)
2046 switch (status) {
2047 case CPL_ERR_RTX_NEG_ADVICE:
2048 return "Retransmit timeout";
2049 case CPL_ERR_PERSIST_NEG_ADVICE:
2050 return "Persist timeout";
2051 case CPL_ERR_KEEPALV_NEG_ADVICE:
2052 return "Keepalive timeout";
2053 default:
2054 return "Unknown";
2058 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2060 ep->snd_win = snd_win;
2061 ep->rcv_win = rcv_win;
2062 pr_debug("snd_win %d rcv_win %d\n",
2063 ep->snd_win, ep->rcv_win);
2066 #define ACT_OPEN_RETRY_COUNT 2
2068 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2069 struct dst_entry *dst, struct c4iw_dev *cdev,
2070 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2072 struct neighbour *n;
2073 int err, step;
2074 struct net_device *pdev;
2076 n = dst_neigh_lookup(dst, peer_ip);
2077 if (!n)
2078 return -ENODEV;
2080 rcu_read_lock();
2081 err = -ENOMEM;
2082 if (n->dev->flags & IFF_LOOPBACK) {
2083 if (iptype == 4)
2084 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2085 else if (IS_ENABLED(CONFIG_IPV6))
2086 for_each_netdev(&init_net, pdev) {
2087 if (ipv6_chk_addr(&init_net,
2088 (struct in6_addr *)peer_ip,
2089 pdev, 1))
2090 break;
2092 else
2093 pdev = NULL;
2095 if (!pdev) {
2096 err = -ENODEV;
2097 goto out;
2099 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2100 n, pdev, rt_tos2priority(tos));
2101 if (!ep->l2t) {
2102 dev_put(pdev);
2103 goto out;
2105 ep->mtu = pdev->mtu;
2106 ep->tx_chan = cxgb4_port_chan(pdev);
2107 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2108 step = cdev->rdev.lldi.ntxq /
2109 cdev->rdev.lldi.nchan;
2110 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2111 step = cdev->rdev.lldi.nrxq /
2112 cdev->rdev.lldi.nchan;
2113 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2114 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2115 cxgb4_port_idx(pdev) * step];
2116 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2117 dev_put(pdev);
2118 } else {
2119 pdev = get_real_dev(n->dev);
2120 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2121 n, pdev, rt_tos2priority(tos));
2122 if (!ep->l2t)
2123 goto out;
2124 ep->mtu = dst_mtu(dst);
2125 ep->tx_chan = cxgb4_port_chan(pdev);
2126 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2127 step = cdev->rdev.lldi.ntxq /
2128 cdev->rdev.lldi.nchan;
2129 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2130 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2131 step = cdev->rdev.lldi.nrxq /
2132 cdev->rdev.lldi.nchan;
2133 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2134 cxgb4_port_idx(pdev) * step];
2135 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2137 if (clear_mpa_v1) {
2138 ep->retry_with_mpa_v1 = 0;
2139 ep->tried_with_mpa_v1 = 0;
2142 err = 0;
2143 out:
2144 rcu_read_unlock();
2146 neigh_release(n);
2148 return err;
2151 static int c4iw_reconnect(struct c4iw_ep *ep)
2153 int err = 0;
2154 int size = 0;
2155 struct sockaddr_in *laddr = (struct sockaddr_in *)
2156 &ep->com.cm_id->m_local_addr;
2157 struct sockaddr_in *raddr = (struct sockaddr_in *)
2158 &ep->com.cm_id->m_remote_addr;
2159 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2160 &ep->com.cm_id->m_local_addr;
2161 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2162 &ep->com.cm_id->m_remote_addr;
2163 int iptype;
2164 __u8 *ra;
2166 pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id);
2167 c4iw_init_wr_wait(ep->com.wr_waitp);
2169 /* When MPA revision is different on nodes, the node with MPA_rev=2
2170 * tries to reconnect with MPA_rev 1 for the same EP through
2171 * c4iw_reconnect(), where the same EP is assigned with new tid for
2172 * further connection establishment. As we are using the same EP pointer
2173 * for reconnect, few skbs are used during the previous c4iw_connect(),
2174 * which leaves the EP with inadequate skbs for further
2175 * c4iw_reconnect(), Further causing a crash due to an empty
2176 * skb_list() during peer_abort(). Allocate skbs which is already used.
2178 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2179 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2180 err = -ENOMEM;
2181 goto fail1;
2185 * Allocate an active TID to initiate a TCP connection.
2187 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2188 if (ep->atid == -1) {
2189 pr_err("%s - cannot alloc atid\n", __func__);
2190 err = -ENOMEM;
2191 goto fail2;
2193 err = xa_insert_irq(&ep->com.dev->atids, ep->atid, ep, GFP_KERNEL);
2194 if (err)
2195 goto fail2a;
2197 /* find a route */
2198 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2199 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2200 laddr->sin_addr.s_addr,
2201 raddr->sin_addr.s_addr,
2202 laddr->sin_port,
2203 raddr->sin_port, ep->com.cm_id->tos);
2204 iptype = 4;
2205 ra = (__u8 *)&raddr->sin_addr;
2206 } else {
2207 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2208 get_real_dev,
2209 laddr6->sin6_addr.s6_addr,
2210 raddr6->sin6_addr.s6_addr,
2211 laddr6->sin6_port,
2212 raddr6->sin6_port,
2213 ep->com.cm_id->tos,
2214 raddr6->sin6_scope_id);
2215 iptype = 6;
2216 ra = (__u8 *)&raddr6->sin6_addr;
2218 if (!ep->dst) {
2219 pr_err("%s - cannot find route\n", __func__);
2220 err = -EHOSTUNREACH;
2221 goto fail3;
2223 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2224 ep->com.dev->rdev.lldi.adapter_type,
2225 ep->com.cm_id->tos);
2226 if (err) {
2227 pr_err("%s - cannot alloc l2e\n", __func__);
2228 goto fail4;
2231 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2232 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2233 ep->l2t->idx);
2235 state_set(&ep->com, CONNECTING);
2236 ep->tos = ep->com.cm_id->tos;
2238 /* send connect request to rnic */
2239 err = send_connect(ep);
2240 if (!err)
2241 goto out;
2243 cxgb4_l2t_release(ep->l2t);
2244 fail4:
2245 dst_release(ep->dst);
2246 fail3:
2247 xa_erase_irq(&ep->com.dev->atids, ep->atid);
2248 fail2a:
2249 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2250 fail2:
2252 * remember to send notification to upper layer.
2253 * We are in here so the upper layer is not aware that this is
2254 * re-connect attempt and so, upper layer is still waiting for
2255 * response of 1st connect request.
2257 connect_reply_upcall(ep, -ECONNRESET);
2258 fail1:
2259 c4iw_put_ep(&ep->com);
2260 out:
2261 return err;
2264 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2266 struct c4iw_ep *ep;
2267 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2268 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2269 ntohl(rpl->atid_status)));
2270 struct tid_info *t = dev->rdev.lldi.tids;
2271 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2272 struct sockaddr_in *la;
2273 struct sockaddr_in *ra;
2274 struct sockaddr_in6 *la6;
2275 struct sockaddr_in6 *ra6;
2276 int ret = 0;
2278 ep = lookup_atid(t, atid);
2279 la = (struct sockaddr_in *)&ep->com.local_addr;
2280 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2281 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2282 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2284 pr_debug("ep %p atid %u status %u errno %d\n", ep, atid,
2285 status, status2errno(status));
2287 if (cxgb_is_neg_adv(status)) {
2288 pr_debug("Connection problems for atid %u status %u (%s)\n",
2289 atid, status, neg_adv_str(status));
2290 ep->stats.connect_neg_adv++;
2291 mutex_lock(&dev->rdev.stats.lock);
2292 dev->rdev.stats.neg_adv++;
2293 mutex_unlock(&dev->rdev.stats.lock);
2294 return 0;
2297 set_bit(ACT_OPEN_RPL, &ep->com.history);
2300 * Log interesting failures.
2302 switch (status) {
2303 case CPL_ERR_CONN_RESET:
2304 case CPL_ERR_CONN_TIMEDOUT:
2305 break;
2306 case CPL_ERR_TCAM_FULL:
2307 mutex_lock(&dev->rdev.stats.lock);
2308 dev->rdev.stats.tcam_full++;
2309 mutex_unlock(&dev->rdev.stats.lock);
2310 if (ep->com.local_addr.ss_family == AF_INET &&
2311 dev->rdev.lldi.enable_fw_ofld_conn) {
2312 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2313 ntohl(rpl->atid_status))));
2314 if (ret)
2315 goto fail;
2316 return 0;
2318 break;
2319 case CPL_ERR_CONN_EXIST:
2320 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2321 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2322 if (ep->com.remote_addr.ss_family == AF_INET6) {
2323 struct sockaddr_in6 *sin6 =
2324 (struct sockaddr_in6 *)
2325 &ep->com.local_addr;
2326 cxgb4_clip_release(
2327 ep->com.dev->rdev.lldi.ports[0],
2328 (const u32 *)
2329 &sin6->sin6_addr.s6_addr, 1);
2331 xa_erase_irq(&ep->com.dev->atids, atid);
2332 cxgb4_free_atid(t, atid);
2333 dst_release(ep->dst);
2334 cxgb4_l2t_release(ep->l2t);
2335 c4iw_reconnect(ep);
2336 return 0;
2338 break;
2339 default:
2340 if (ep->com.local_addr.ss_family == AF_INET) {
2341 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2342 atid, status, status2errno(status),
2343 &la->sin_addr.s_addr, ntohs(la->sin_port),
2344 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2345 } else {
2346 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2347 atid, status, status2errno(status),
2348 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2349 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2351 break;
2354 fail:
2355 connect_reply_upcall(ep, status2errno(status));
2356 state_set(&ep->com, DEAD);
2358 if (ep->com.remote_addr.ss_family == AF_INET6) {
2359 struct sockaddr_in6 *sin6 =
2360 (struct sockaddr_in6 *)&ep->com.local_addr;
2361 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2362 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2364 if (status && act_open_has_tid(status))
2365 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2366 ep->com.local_addr.ss_family);
2368 xa_erase_irq(&ep->com.dev->atids, atid);
2369 cxgb4_free_atid(t, atid);
2370 dst_release(ep->dst);
2371 cxgb4_l2t_release(ep->l2t);
2372 c4iw_put_ep(&ep->com);
2374 return 0;
2377 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2379 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2380 unsigned int stid = GET_TID(rpl);
2381 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2383 if (!ep) {
2384 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2385 goto out;
2387 pr_debug("ep %p status %d error %d\n", ep,
2388 rpl->status, status2errno(rpl->status));
2389 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2390 c4iw_put_ep(&ep->com);
2391 out:
2392 return 0;
2395 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2397 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2398 unsigned int stid = GET_TID(rpl);
2399 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2401 if (!ep) {
2402 pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2403 goto out;
2405 pr_debug("ep %p\n", ep);
2406 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2407 c4iw_put_ep(&ep->com);
2408 out:
2409 return 0;
2412 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2413 struct cpl_pass_accept_req *req)
2415 struct cpl_pass_accept_rpl *rpl;
2416 unsigned int mtu_idx;
2417 u64 opt0;
2418 u32 opt2;
2419 u32 wscale;
2420 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2421 int win;
2422 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2424 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2425 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2426 enable_tcp_timestamps && req->tcpopt.tstamp,
2427 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2428 wscale = cxgb_compute_wscale(rcv_win);
2431 * Specify the largest window that will fit in opt0. The
2432 * remainder will be specified in the rx_data_ack.
2434 win = ep->rcv_win >> 10;
2435 if (win > RCV_BUFSIZ_M)
2436 win = RCV_BUFSIZ_M;
2437 opt0 = (nocong ? NO_CONG_F : 0) |
2438 KEEP_ALIVE_F |
2439 DELACK_F |
2440 WND_SCALE_V(wscale) |
2441 MSS_IDX_V(mtu_idx) |
2442 L2T_IDX_V(ep->l2t->idx) |
2443 TX_CHAN_V(ep->tx_chan) |
2444 SMAC_SEL_V(ep->smac_idx) |
2445 DSCP_V(ep->tos >> 2) |
2446 ULP_MODE_V(ULP_MODE_TCPDDP) |
2447 RCV_BUFSIZ_V(win);
2448 opt2 = RX_CHANNEL_V(0) |
2449 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2451 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2452 opt2 |= TSTAMPS_EN_F;
2453 if (enable_tcp_sack && req->tcpopt.sack)
2454 opt2 |= SACK_EN_F;
2455 if (wscale && enable_tcp_window_scaling)
2456 opt2 |= WND_SCALE_EN_F;
2457 if (enable_ecn) {
2458 const struct tcphdr *tcph;
2459 u32 hlen = ntohl(req->hdr_len);
2461 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2462 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2463 IP_HDR_LEN_G(hlen);
2464 else
2465 tcph = (const void *)(req + 1) +
2466 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2467 if (tcph->ece && tcph->cwr)
2468 opt2 |= CCTRL_ECN_V(1);
2471 skb_get(skb);
2472 rpl = cplhdr(skb);
2473 if (!is_t4(adapter_type)) {
2474 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2475 rpl5 = (void *)rpl;
2476 INIT_TP_WR(rpl5, ep->hwtid);
2477 } else {
2478 skb_trim(skb, sizeof(*rpl));
2479 INIT_TP_WR(rpl, ep->hwtid);
2481 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2482 ep->hwtid));
2484 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2485 u32 isn = (prandom_u32() & ~7UL) - 1;
2486 opt2 |= T5_OPT_2_VALID_F;
2487 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2488 opt2 |= T5_ISS_F;
2489 rpl5 = (void *)rpl;
2490 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2491 if (peer2peer)
2492 isn += 4;
2493 rpl5->iss = cpu_to_be32(isn);
2494 pr_debug("iss %u\n", be32_to_cpu(rpl5->iss));
2497 rpl->opt0 = cpu_to_be64(opt0);
2498 rpl->opt2 = cpu_to_be32(opt2);
2499 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2500 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2502 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2505 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2507 pr_debug("c4iw_dev %p tid %u\n", dev, hwtid);
2508 skb_trim(skb, sizeof(struct cpl_tid_release));
2509 release_tid(&dev->rdev, hwtid, skb);
2510 return;
2513 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2515 struct c4iw_ep *child_ep = NULL, *parent_ep;
2516 struct cpl_pass_accept_req *req = cplhdr(skb);
2517 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2518 struct tid_info *t = dev->rdev.lldi.tids;
2519 unsigned int hwtid = GET_TID(req);
2520 struct dst_entry *dst;
2521 __u8 local_ip[16], peer_ip[16];
2522 __be16 local_port, peer_port;
2523 struct sockaddr_in6 *sin6;
2524 int err;
2525 u16 peer_mss = ntohs(req->tcpopt.mss);
2526 int iptype;
2527 unsigned short hdrs;
2528 u8 tos;
2530 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2531 if (!parent_ep) {
2532 pr_err("%s connect request on invalid stid %d\n",
2533 __func__, stid);
2534 goto reject;
2537 if (state_read(&parent_ep->com) != LISTEN) {
2538 pr_err("%s - listening ep not in LISTEN\n", __func__);
2539 goto reject;
2542 if (parent_ep->com.cm_id->tos_set)
2543 tos = parent_ep->com.cm_id->tos;
2544 else
2545 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2547 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2548 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2550 /* Find output route */
2551 if (iptype == 4) {
2552 pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2553 , parent_ep, hwtid,
2554 local_ip, peer_ip, ntohs(local_port),
2555 ntohs(peer_port), peer_mss);
2556 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2557 *(__be32 *)local_ip, *(__be32 *)peer_ip,
2558 local_port, peer_port, tos);
2559 } else {
2560 pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2561 , parent_ep, hwtid,
2562 local_ip, peer_ip, ntohs(local_port),
2563 ntohs(peer_port), peer_mss);
2564 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2565 local_ip, peer_ip, local_port, peer_port,
2566 tos,
2567 ((struct sockaddr_in6 *)
2568 &parent_ep->com.local_addr)->sin6_scope_id);
2570 if (!dst) {
2571 pr_err("%s - failed to find dst entry!\n", __func__);
2572 goto reject;
2575 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2576 if (!child_ep) {
2577 pr_err("%s - failed to allocate ep entry!\n", __func__);
2578 dst_release(dst);
2579 goto reject;
2582 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2583 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2584 if (err) {
2585 pr_err("%s - failed to allocate l2t entry!\n", __func__);
2586 dst_release(dst);
2587 kfree(child_ep);
2588 goto reject;
2591 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2592 sizeof(struct tcphdr) +
2593 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2594 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2595 child_ep->mtu = peer_mss + hdrs;
2597 skb_queue_head_init(&child_ep->com.ep_skb_list);
2598 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2599 goto fail;
2601 state_set(&child_ep->com, CONNECTING);
2602 child_ep->com.dev = dev;
2603 child_ep->com.cm_id = NULL;
2605 if (iptype == 4) {
2606 struct sockaddr_in *sin = (struct sockaddr_in *)
2607 &child_ep->com.local_addr;
2609 sin->sin_family = AF_INET;
2610 sin->sin_port = local_port;
2611 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2613 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2614 sin->sin_family = AF_INET;
2615 sin->sin_port = ((struct sockaddr_in *)
2616 &parent_ep->com.local_addr)->sin_port;
2617 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2619 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2620 sin->sin_family = AF_INET;
2621 sin->sin_port = peer_port;
2622 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2623 } else {
2624 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2625 sin6->sin6_family = PF_INET6;
2626 sin6->sin6_port = local_port;
2627 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2629 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2630 sin6->sin6_family = PF_INET6;
2631 sin6->sin6_port = ((struct sockaddr_in6 *)
2632 &parent_ep->com.local_addr)->sin6_port;
2633 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2635 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2636 sin6->sin6_family = PF_INET6;
2637 sin6->sin6_port = peer_port;
2638 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2641 c4iw_get_ep(&parent_ep->com);
2642 child_ep->parent_ep = parent_ep;
2643 child_ep->tos = tos;
2644 child_ep->dst = dst;
2645 child_ep->hwtid = hwtid;
2647 pr_debug("tx_chan %u smac_idx %u rss_qid %u\n",
2648 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2650 timer_setup(&child_ep->timer, ep_timeout, 0);
2651 cxgb4_insert_tid(t, child_ep, hwtid,
2652 child_ep->com.local_addr.ss_family);
2653 insert_ep_tid(child_ep);
2654 if (accept_cr(child_ep, skb, req)) {
2655 c4iw_put_ep(&parent_ep->com);
2656 release_ep_resources(child_ep);
2657 } else {
2658 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2660 if (iptype == 6) {
2661 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2662 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2663 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2665 goto out;
2666 fail:
2667 c4iw_put_ep(&child_ep->com);
2668 reject:
2669 reject_cr(dev, hwtid, skb);
2670 out:
2671 if (parent_ep)
2672 c4iw_put_ep(&parent_ep->com);
2673 return 0;
2676 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2678 struct c4iw_ep *ep;
2679 struct cpl_pass_establish *req = cplhdr(skb);
2680 unsigned int tid = GET_TID(req);
2681 int ret;
2682 u16 tcp_opt = ntohs(req->tcp_opt);
2684 ep = get_ep_from_tid(dev, tid);
2685 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2686 ep->snd_seq = be32_to_cpu(req->snd_isn);
2687 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2688 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
2690 pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid, tcp_opt);
2692 set_emss(ep, tcp_opt);
2694 dst_confirm(ep->dst);
2695 mutex_lock(&ep->com.mutex);
2696 ep->com.state = MPA_REQ_WAIT;
2697 start_ep_timer(ep);
2698 set_bit(PASS_ESTAB, &ep->com.history);
2699 ret = send_flowc(ep);
2700 mutex_unlock(&ep->com.mutex);
2701 if (ret)
2702 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2703 c4iw_put_ep(&ep->com);
2705 return 0;
2708 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2710 struct cpl_peer_close *hdr = cplhdr(skb);
2711 struct c4iw_ep *ep;
2712 struct c4iw_qp_attributes attrs;
2713 int disconnect = 1;
2714 int release = 0;
2715 unsigned int tid = GET_TID(hdr);
2716 int ret;
2718 ep = get_ep_from_tid(dev, tid);
2719 if (!ep)
2720 return 0;
2722 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2723 dst_confirm(ep->dst);
2725 set_bit(PEER_CLOSE, &ep->com.history);
2726 mutex_lock(&ep->com.mutex);
2727 switch (ep->com.state) {
2728 case MPA_REQ_WAIT:
2729 __state_set(&ep->com, CLOSING);
2730 break;
2731 case MPA_REQ_SENT:
2732 __state_set(&ep->com, CLOSING);
2733 connect_reply_upcall(ep, -ECONNRESET);
2734 break;
2735 case MPA_REQ_RCVD:
2738 * We're gonna mark this puppy DEAD, but keep
2739 * the reference on it until the ULP accepts or
2740 * rejects the CR. Also wake up anyone waiting
2741 * in rdma connection migration (see c4iw_accept_cr()).
2743 __state_set(&ep->com, CLOSING);
2744 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2745 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2746 break;
2747 case MPA_REP_SENT:
2748 __state_set(&ep->com, CLOSING);
2749 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2750 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2751 break;
2752 case FPDU_MODE:
2753 start_ep_timer(ep);
2754 __state_set(&ep->com, CLOSING);
2755 attrs.next_state = C4IW_QP_STATE_CLOSING;
2756 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2757 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2758 if (ret != -ECONNRESET) {
2759 peer_close_upcall(ep);
2760 disconnect = 1;
2762 break;
2763 case ABORTING:
2764 disconnect = 0;
2765 break;
2766 case CLOSING:
2767 __state_set(&ep->com, MORIBUND);
2768 disconnect = 0;
2769 break;
2770 case MORIBUND:
2771 (void)stop_ep_timer(ep);
2772 if (ep->com.cm_id && ep->com.qp) {
2773 attrs.next_state = C4IW_QP_STATE_IDLE;
2774 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2775 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2777 close_complete_upcall(ep, 0);
2778 __state_set(&ep->com, DEAD);
2779 release = 1;
2780 disconnect = 0;
2781 break;
2782 case DEAD:
2783 disconnect = 0;
2784 break;
2785 default:
2786 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2788 mutex_unlock(&ep->com.mutex);
2789 if (disconnect)
2790 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2791 if (release)
2792 release_ep_resources(ep);
2793 c4iw_put_ep(&ep->com);
2794 return 0;
2797 static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep)
2799 complete_cached_srq_buffers(ep, ep->srqe_idx);
2800 if (ep->com.cm_id && ep->com.qp) {
2801 struct c4iw_qp_attributes attrs;
2803 attrs.next_state = C4IW_QP_STATE_ERROR;
2804 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2805 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2807 peer_abort_upcall(ep);
2808 release_ep_resources(ep);
2809 c4iw_put_ep(&ep->com);
2812 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2814 struct cpl_abort_req_rss6 *req = cplhdr(skb);
2815 struct c4iw_ep *ep;
2816 struct sk_buff *rpl_skb;
2817 struct c4iw_qp_attributes attrs;
2818 int ret;
2819 int release = 0;
2820 unsigned int tid = GET_TID(req);
2821 u8 status;
2822 u32 srqidx;
2824 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2826 ep = get_ep_from_tid(dev, tid);
2827 if (!ep)
2828 return 0;
2830 status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status));
2832 if (cxgb_is_neg_adv(status)) {
2833 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
2834 ep->hwtid, status, neg_adv_str(status));
2835 ep->stats.abort_neg_adv++;
2836 mutex_lock(&dev->rdev.stats.lock);
2837 dev->rdev.stats.neg_adv++;
2838 mutex_unlock(&dev->rdev.stats.lock);
2839 goto deref_ep;
2842 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
2843 ep->com.state);
2844 set_bit(PEER_ABORT, &ep->com.history);
2847 * Wake up any threads in rdma_init() or rdma_fini().
2848 * However, this is not needed if com state is just
2849 * MPA_REQ_SENT
2851 if (ep->com.state != MPA_REQ_SENT)
2852 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2854 mutex_lock(&ep->com.mutex);
2855 switch (ep->com.state) {
2856 case CONNECTING:
2857 c4iw_put_ep(&ep->parent_ep->com);
2858 break;
2859 case MPA_REQ_WAIT:
2860 (void)stop_ep_timer(ep);
2861 break;
2862 case MPA_REQ_SENT:
2863 (void)stop_ep_timer(ep);
2864 if (status != CPL_ERR_CONN_RESET || mpa_rev == 1 ||
2865 (mpa_rev == 2 && ep->tried_with_mpa_v1))
2866 connect_reply_upcall(ep, -ECONNRESET);
2867 else {
2869 * we just don't send notification upwards because we
2870 * want to retry with mpa_v1 without upper layers even
2871 * knowing it.
2873 * do some housekeeping so as to re-initiate the
2874 * connection
2876 pr_info("%s: mpa_rev=%d. Retrying with mpav1\n",
2877 __func__, mpa_rev);
2878 ep->retry_with_mpa_v1 = 1;
2880 break;
2881 case MPA_REP_SENT:
2882 break;
2883 case MPA_REQ_RCVD:
2884 break;
2885 case MORIBUND:
2886 case CLOSING:
2887 stop_ep_timer(ep);
2888 fallthrough;
2889 case FPDU_MODE:
2890 if (ep->com.qp && ep->com.qp->srq) {
2891 srqidx = ABORT_RSS_SRQIDX_G(
2892 be32_to_cpu(req->srqidx_status));
2893 if (srqidx) {
2894 complete_cached_srq_buffers(ep, srqidx);
2895 } else {
2896 /* Hold ep ref until finish_peer_abort() */
2897 c4iw_get_ep(&ep->com);
2898 __state_set(&ep->com, ABORTING);
2899 set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags);
2900 read_tcb(ep);
2901 break;
2906 if (ep->com.cm_id && ep->com.qp) {
2907 attrs.next_state = C4IW_QP_STATE_ERROR;
2908 ret = c4iw_modify_qp(ep->com.qp->rhp,
2909 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2910 &attrs, 1);
2911 if (ret)
2912 pr_err("%s - qp <- error failed!\n", __func__);
2914 peer_abort_upcall(ep);
2915 break;
2916 case ABORTING:
2917 break;
2918 case DEAD:
2919 pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2920 mutex_unlock(&ep->com.mutex);
2921 goto deref_ep;
2922 default:
2923 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2924 break;
2926 dst_confirm(ep->dst);
2927 if (ep->com.state != ABORTING) {
2928 __state_set(&ep->com, DEAD);
2929 /* we don't release if we want to retry with mpa_v1 */
2930 if (!ep->retry_with_mpa_v1)
2931 release = 1;
2933 mutex_unlock(&ep->com.mutex);
2935 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2936 if (WARN_ON(!rpl_skb)) {
2937 release = 1;
2938 goto out;
2941 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2943 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2944 out:
2945 if (release)
2946 release_ep_resources(ep);
2947 else if (ep->retry_with_mpa_v1) {
2948 if (ep->com.remote_addr.ss_family == AF_INET6) {
2949 struct sockaddr_in6 *sin6 =
2950 (struct sockaddr_in6 *)
2951 &ep->com.local_addr;
2952 cxgb4_clip_release(
2953 ep->com.dev->rdev.lldi.ports[0],
2954 (const u32 *)&sin6->sin6_addr.s6_addr,
2957 xa_erase_irq(&ep->com.dev->hwtids, ep->hwtid);
2958 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2959 ep->com.local_addr.ss_family);
2960 dst_release(ep->dst);
2961 cxgb4_l2t_release(ep->l2t);
2962 c4iw_reconnect(ep);
2965 deref_ep:
2966 c4iw_put_ep(&ep->com);
2967 /* Dereferencing ep, referenced in peer_abort_intr() */
2968 c4iw_put_ep(&ep->com);
2969 return 0;
2972 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2974 struct c4iw_ep *ep;
2975 struct c4iw_qp_attributes attrs;
2976 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2977 int release = 0;
2978 unsigned int tid = GET_TID(rpl);
2980 ep = get_ep_from_tid(dev, tid);
2981 if (!ep)
2982 return 0;
2984 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2986 /* The cm_id may be null if we failed to connect */
2987 mutex_lock(&ep->com.mutex);
2988 set_bit(CLOSE_CON_RPL, &ep->com.history);
2989 switch (ep->com.state) {
2990 case CLOSING:
2991 __state_set(&ep->com, MORIBUND);
2992 break;
2993 case MORIBUND:
2994 (void)stop_ep_timer(ep);
2995 if ((ep->com.cm_id) && (ep->com.qp)) {
2996 attrs.next_state = C4IW_QP_STATE_IDLE;
2997 c4iw_modify_qp(ep->com.qp->rhp,
2998 ep->com.qp,
2999 C4IW_QP_ATTR_NEXT_STATE,
3000 &attrs, 1);
3002 close_complete_upcall(ep, 0);
3003 __state_set(&ep->com, DEAD);
3004 release = 1;
3005 break;
3006 case ABORTING:
3007 case DEAD:
3008 break;
3009 default:
3010 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3011 break;
3013 mutex_unlock(&ep->com.mutex);
3014 if (release)
3015 release_ep_resources(ep);
3016 c4iw_put_ep(&ep->com);
3017 return 0;
3020 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3022 struct cpl_rdma_terminate *rpl = cplhdr(skb);
3023 unsigned int tid = GET_TID(rpl);
3024 struct c4iw_ep *ep;
3025 struct c4iw_qp_attributes attrs;
3027 ep = get_ep_from_tid(dev, tid);
3029 if (ep) {
3030 if (ep->com.qp) {
3031 pr_warn("TERM received tid %u qpid %u\n", tid,
3032 ep->com.qp->wq.sq.qid);
3033 attrs.next_state = C4IW_QP_STATE_TERMINATE;
3034 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3035 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3038 /* As per draft-hilland-iwarp-verbs-v1.0, sec 6.2.3,
3039 * when entering the TERM state the RNIC MUST initiate a CLOSE.
3041 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3042 c4iw_put_ep(&ep->com);
3043 } else
3044 pr_warn("TERM received tid %u no ep/qp\n", tid);
3046 return 0;
3050 * Upcall from the adapter indicating data has been transmitted.
3051 * For us its just the single MPA request or reply. We can now free
3052 * the skb holding the mpa message.
3054 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3056 struct c4iw_ep *ep;
3057 struct cpl_fw4_ack *hdr = cplhdr(skb);
3058 u8 credits = hdr->credits;
3059 unsigned int tid = GET_TID(hdr);
3062 ep = get_ep_from_tid(dev, tid);
3063 if (!ep)
3064 return 0;
3065 pr_debug("ep %p tid %u credits %u\n",
3066 ep, ep->hwtid, credits);
3067 if (credits == 0) {
3068 pr_debug("0 credit ack ep %p tid %u state %u\n",
3069 ep, ep->hwtid, state_read(&ep->com));
3070 goto out;
3073 dst_confirm(ep->dst);
3074 if (ep->mpa_skb) {
3075 pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
3076 ep, ep->hwtid, state_read(&ep->com),
3077 ep->mpa_attr.initiator ? 1 : 0);
3078 mutex_lock(&ep->com.mutex);
3079 kfree_skb(ep->mpa_skb);
3080 ep->mpa_skb = NULL;
3081 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3082 stop_ep_timer(ep);
3083 mutex_unlock(&ep->com.mutex);
3085 out:
3086 c4iw_put_ep(&ep->com);
3087 return 0;
3090 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3092 int abort;
3093 struct c4iw_ep *ep = to_ep(cm_id);
3095 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3097 mutex_lock(&ep->com.mutex);
3098 if (ep->com.state != MPA_REQ_RCVD) {
3099 mutex_unlock(&ep->com.mutex);
3100 c4iw_put_ep(&ep->com);
3101 return -ECONNRESET;
3103 set_bit(ULP_REJECT, &ep->com.history);
3104 if (mpa_rev == 0)
3105 abort = 1;
3106 else
3107 abort = send_mpa_reject(ep, pdata, pdata_len);
3108 mutex_unlock(&ep->com.mutex);
3110 stop_ep_timer(ep);
3111 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3112 c4iw_put_ep(&ep->com);
3113 return 0;
3116 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3118 int err;
3119 struct c4iw_qp_attributes attrs;
3120 enum c4iw_qp_attr_mask mask;
3121 struct c4iw_ep *ep = to_ep(cm_id);
3122 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3123 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3124 int abort = 0;
3126 pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3128 mutex_lock(&ep->com.mutex);
3129 if (ep->com.state != MPA_REQ_RCVD) {
3130 err = -ECONNRESET;
3131 goto err_out;
3134 if (!qp) {
3135 err = -EINVAL;
3136 goto err_out;
3139 set_bit(ULP_ACCEPT, &ep->com.history);
3140 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3141 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3142 err = -EINVAL;
3143 goto err_abort;
3146 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3147 if (conn_param->ord > ep->ird) {
3148 if (RELAXED_IRD_NEGOTIATION) {
3149 conn_param->ord = ep->ird;
3150 } else {
3151 ep->ird = conn_param->ird;
3152 ep->ord = conn_param->ord;
3153 send_mpa_reject(ep, conn_param->private_data,
3154 conn_param->private_data_len);
3155 err = -ENOMEM;
3156 goto err_abort;
3159 if (conn_param->ird < ep->ord) {
3160 if (RELAXED_IRD_NEGOTIATION &&
3161 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3162 conn_param->ird = ep->ord;
3163 } else {
3164 err = -ENOMEM;
3165 goto err_abort;
3169 ep->ird = conn_param->ird;
3170 ep->ord = conn_param->ord;
3172 if (ep->mpa_attr.version == 1) {
3173 if (peer2peer && ep->ird == 0)
3174 ep->ird = 1;
3175 } else {
3176 if (peer2peer &&
3177 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3178 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3179 ep->ird = 1;
3182 pr_debug("ird %d ord %d\n", ep->ird, ep->ord);
3184 ep->com.cm_id = cm_id;
3185 ref_cm_id(&ep->com);
3186 ep->com.qp = qp;
3187 ref_qp(ep);
3189 /* bind QP to EP and move to RTS */
3190 attrs.mpa_attr = ep->mpa_attr;
3191 attrs.max_ird = ep->ird;
3192 attrs.max_ord = ep->ord;
3193 attrs.llp_stream_handle = ep;
3194 attrs.next_state = C4IW_QP_STATE_RTS;
3196 /* bind QP and TID with INIT_WR */
3197 mask = C4IW_QP_ATTR_NEXT_STATE |
3198 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3199 C4IW_QP_ATTR_MPA_ATTR |
3200 C4IW_QP_ATTR_MAX_IRD |
3201 C4IW_QP_ATTR_MAX_ORD;
3203 err = c4iw_modify_qp(ep->com.qp->rhp,
3204 ep->com.qp, mask, &attrs, 1);
3205 if (err)
3206 goto err_deref_cm_id;
3208 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3209 err = send_mpa_reply(ep, conn_param->private_data,
3210 conn_param->private_data_len);
3211 if (err)
3212 goto err_deref_cm_id;
3214 __state_set(&ep->com, FPDU_MODE);
3215 established_upcall(ep);
3216 mutex_unlock(&ep->com.mutex);
3217 c4iw_put_ep(&ep->com);
3218 return 0;
3219 err_deref_cm_id:
3220 deref_cm_id(&ep->com);
3221 err_abort:
3222 abort = 1;
3223 err_out:
3224 mutex_unlock(&ep->com.mutex);
3225 if (abort)
3226 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3227 c4iw_put_ep(&ep->com);
3228 return err;
3231 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3233 struct in_device *ind;
3234 int found = 0;
3235 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3236 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3237 const struct in_ifaddr *ifa;
3239 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3240 if (!ind)
3241 return -EADDRNOTAVAIL;
3242 rcu_read_lock();
3243 in_dev_for_each_ifa_rcu(ifa, ind) {
3244 if (ifa->ifa_flags & IFA_F_SECONDARY)
3245 continue;
3246 laddr->sin_addr.s_addr = ifa->ifa_address;
3247 raddr->sin_addr.s_addr = ifa->ifa_address;
3248 found = 1;
3249 break;
3251 rcu_read_unlock();
3253 in_dev_put(ind);
3254 return found ? 0 : -EADDRNOTAVAIL;
3257 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3258 unsigned char banned_flags)
3260 struct inet6_dev *idev;
3261 int err = -EADDRNOTAVAIL;
3263 rcu_read_lock();
3264 idev = __in6_dev_get(dev);
3265 if (idev != NULL) {
3266 struct inet6_ifaddr *ifp;
3268 read_lock_bh(&idev->lock);
3269 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3270 if (ifp->scope == IFA_LINK &&
3271 !(ifp->flags & banned_flags)) {
3272 memcpy(addr, &ifp->addr, 16);
3273 err = 0;
3274 break;
3277 read_unlock_bh(&idev->lock);
3279 rcu_read_unlock();
3280 return err;
3283 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3285 struct in6_addr addr;
3286 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3287 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3289 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3290 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3291 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3292 return 0;
3294 return -EADDRNOTAVAIL;
3297 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3299 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3300 struct c4iw_ep *ep;
3301 int err = 0;
3302 struct sockaddr_in *laddr;
3303 struct sockaddr_in *raddr;
3304 struct sockaddr_in6 *laddr6;
3305 struct sockaddr_in6 *raddr6;
3306 __u8 *ra;
3307 int iptype;
3309 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3310 (conn_param->ird > cur_max_read_depth(dev))) {
3311 err = -EINVAL;
3312 goto out;
3314 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3315 if (!ep) {
3316 pr_err("%s - cannot alloc ep\n", __func__);
3317 err = -ENOMEM;
3318 goto out;
3321 skb_queue_head_init(&ep->com.ep_skb_list);
3322 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3323 err = -ENOMEM;
3324 goto fail1;
3327 timer_setup(&ep->timer, ep_timeout, 0);
3328 ep->plen = conn_param->private_data_len;
3329 if (ep->plen)
3330 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3331 conn_param->private_data, ep->plen);
3332 ep->ird = conn_param->ird;
3333 ep->ord = conn_param->ord;
3335 if (peer2peer && ep->ord == 0)
3336 ep->ord = 1;
3338 ep->com.cm_id = cm_id;
3339 ref_cm_id(&ep->com);
3340 cm_id->provider_data = ep;
3341 ep->com.dev = dev;
3342 ep->com.qp = get_qhp(dev, conn_param->qpn);
3343 if (!ep->com.qp) {
3344 pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3345 err = -EINVAL;
3346 goto fail2;
3348 ref_qp(ep);
3349 pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn,
3350 ep->com.qp, cm_id);
3353 * Allocate an active TID to initiate a TCP connection.
3355 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3356 if (ep->atid == -1) {
3357 pr_err("%s - cannot alloc atid\n", __func__);
3358 err = -ENOMEM;
3359 goto fail2;
3361 err = xa_insert_irq(&dev->atids, ep->atid, ep, GFP_KERNEL);
3362 if (err)
3363 goto fail5;
3365 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3366 sizeof(ep->com.local_addr));
3367 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3368 sizeof(ep->com.remote_addr));
3370 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3371 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3372 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3373 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3375 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3376 iptype = 4;
3377 ra = (__u8 *)&raddr->sin_addr;
3380 * Handle loopback requests to INADDR_ANY.
3382 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3383 err = pick_local_ipaddrs(dev, cm_id);
3384 if (err)
3385 goto fail3;
3388 /* find a route */
3389 pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3390 &laddr->sin_addr, ntohs(laddr->sin_port),
3391 ra, ntohs(raddr->sin_port));
3392 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3393 laddr->sin_addr.s_addr,
3394 raddr->sin_addr.s_addr,
3395 laddr->sin_port,
3396 raddr->sin_port, cm_id->tos);
3397 } else {
3398 iptype = 6;
3399 ra = (__u8 *)&raddr6->sin6_addr;
3402 * Handle loopback requests to INADDR_ANY.
3404 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3405 err = pick_local_ip6addrs(dev, cm_id);
3406 if (err)
3407 goto fail3;
3410 /* find a route */
3411 pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3412 laddr6->sin6_addr.s6_addr,
3413 ntohs(laddr6->sin6_port),
3414 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3415 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3416 laddr6->sin6_addr.s6_addr,
3417 raddr6->sin6_addr.s6_addr,
3418 laddr6->sin6_port,
3419 raddr6->sin6_port, cm_id->tos,
3420 raddr6->sin6_scope_id);
3422 if (!ep->dst) {
3423 pr_err("%s - cannot find route\n", __func__);
3424 err = -EHOSTUNREACH;
3425 goto fail3;
3428 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3429 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3430 if (err) {
3431 pr_err("%s - cannot alloc l2e\n", __func__);
3432 goto fail4;
3435 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3436 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3437 ep->l2t->idx);
3439 state_set(&ep->com, CONNECTING);
3440 ep->tos = cm_id->tos;
3442 /* send connect request to rnic */
3443 err = send_connect(ep);
3444 if (!err)
3445 goto out;
3447 cxgb4_l2t_release(ep->l2t);
3448 fail4:
3449 dst_release(ep->dst);
3450 fail3:
3451 xa_erase_irq(&ep->com.dev->atids, ep->atid);
3452 fail5:
3453 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3454 fail2:
3455 skb_queue_purge(&ep->com.ep_skb_list);
3456 deref_cm_id(&ep->com);
3457 fail1:
3458 c4iw_put_ep(&ep->com);
3459 out:
3460 return err;
3463 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3465 int err;
3466 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3467 &ep->com.local_addr;
3469 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3470 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3471 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3472 if (err)
3473 return err;
3475 c4iw_init_wr_wait(ep->com.wr_waitp);
3476 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3477 ep->stid, &sin6->sin6_addr,
3478 sin6->sin6_port,
3479 ep->com.dev->rdev.lldi.rxq_ids[0]);
3480 if (!err)
3481 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3482 ep->com.wr_waitp,
3483 0, 0, __func__);
3484 else if (err > 0)
3485 err = net_xmit_errno(err);
3486 if (err) {
3487 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3488 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3489 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3490 err, ep->stid,
3491 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3493 return err;
3496 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3498 int err;
3499 struct sockaddr_in *sin = (struct sockaddr_in *)
3500 &ep->com.local_addr;
3502 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3503 do {
3504 err = cxgb4_create_server_filter(
3505 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3506 sin->sin_addr.s_addr, sin->sin_port, 0,
3507 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3508 if (err == -EBUSY) {
3509 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3510 err = -EIO;
3511 break;
3513 set_current_state(TASK_UNINTERRUPTIBLE);
3514 schedule_timeout(usecs_to_jiffies(100));
3516 } while (err == -EBUSY);
3517 } else {
3518 c4iw_init_wr_wait(ep->com.wr_waitp);
3519 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3520 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3521 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3522 if (!err)
3523 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3524 ep->com.wr_waitp,
3525 0, 0, __func__);
3526 else if (err > 0)
3527 err = net_xmit_errno(err);
3529 if (err)
3530 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3531 , err, ep->stid,
3532 &sin->sin_addr, ntohs(sin->sin_port));
3533 return err;
3536 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3538 int err = 0;
3539 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3540 struct c4iw_listen_ep *ep;
3542 might_sleep();
3544 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3545 if (!ep) {
3546 pr_err("%s - cannot alloc ep\n", __func__);
3547 err = -ENOMEM;
3548 goto fail1;
3550 skb_queue_head_init(&ep->com.ep_skb_list);
3551 pr_debug("ep %p\n", ep);
3552 ep->com.cm_id = cm_id;
3553 ref_cm_id(&ep->com);
3554 ep->com.dev = dev;
3555 ep->backlog = backlog;
3556 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3557 sizeof(ep->com.local_addr));
3560 * Allocate a server TID.
3562 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3563 ep->com.local_addr.ss_family == AF_INET)
3564 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3565 cm_id->m_local_addr.ss_family, ep);
3566 else
3567 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3568 cm_id->m_local_addr.ss_family, ep);
3570 if (ep->stid == -1) {
3571 pr_err("%s - cannot alloc stid\n", __func__);
3572 err = -ENOMEM;
3573 goto fail2;
3575 err = xa_insert_irq(&dev->stids, ep->stid, ep, GFP_KERNEL);
3576 if (err)
3577 goto fail3;
3579 state_set(&ep->com, LISTEN);
3580 if (ep->com.local_addr.ss_family == AF_INET)
3581 err = create_server4(dev, ep);
3582 else
3583 err = create_server6(dev, ep);
3584 if (!err) {
3585 cm_id->provider_data = ep;
3586 goto out;
3588 xa_erase_irq(&ep->com.dev->stids, ep->stid);
3589 fail3:
3590 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3591 ep->com.local_addr.ss_family);
3592 fail2:
3593 deref_cm_id(&ep->com);
3594 c4iw_put_ep(&ep->com);
3595 fail1:
3596 out:
3597 return err;
3600 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3602 int err;
3603 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3605 pr_debug("ep %p\n", ep);
3607 might_sleep();
3608 state_set(&ep->com, DEAD);
3609 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3610 ep->com.local_addr.ss_family == AF_INET) {
3611 err = cxgb4_remove_server_filter(
3612 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3613 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3614 } else {
3615 struct sockaddr_in6 *sin6;
3616 c4iw_init_wr_wait(ep->com.wr_waitp);
3617 err = cxgb4_remove_server(
3618 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3619 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3620 if (err)
3621 goto done;
3622 err = c4iw_wait_for_reply(&ep->com.dev->rdev, ep->com.wr_waitp,
3623 0, 0, __func__);
3624 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3625 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3626 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3628 xa_erase_irq(&ep->com.dev->stids, ep->stid);
3629 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3630 ep->com.local_addr.ss_family);
3631 done:
3632 deref_cm_id(&ep->com);
3633 c4iw_put_ep(&ep->com);
3634 return err;
3637 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3639 int ret = 0;
3640 int close = 0;
3641 int fatal = 0;
3642 struct c4iw_rdev *rdev;
3644 mutex_lock(&ep->com.mutex);
3646 pr_debug("ep %p state %s, abrupt %d\n", ep,
3647 states[ep->com.state], abrupt);
3650 * Ref the ep here in case we have fatal errors causing the
3651 * ep to be released and freed.
3653 c4iw_get_ep(&ep->com);
3655 rdev = &ep->com.dev->rdev;
3656 if (c4iw_fatal_error(rdev)) {
3657 fatal = 1;
3658 close_complete_upcall(ep, -EIO);
3659 ep->com.state = DEAD;
3661 switch (ep->com.state) {
3662 case MPA_REQ_WAIT:
3663 case MPA_REQ_SENT:
3664 case MPA_REQ_RCVD:
3665 case MPA_REP_SENT:
3666 case FPDU_MODE:
3667 case CONNECTING:
3668 close = 1;
3669 if (abrupt)
3670 ep->com.state = ABORTING;
3671 else {
3672 ep->com.state = CLOSING;
3675 * if we close before we see the fw4_ack() then we fix
3676 * up the timer state since we're reusing it.
3678 if (ep->mpa_skb &&
3679 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3680 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3681 stop_ep_timer(ep);
3683 start_ep_timer(ep);
3685 set_bit(CLOSE_SENT, &ep->com.flags);
3686 break;
3687 case CLOSING:
3688 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3689 close = 1;
3690 if (abrupt) {
3691 (void)stop_ep_timer(ep);
3692 ep->com.state = ABORTING;
3693 } else
3694 ep->com.state = MORIBUND;
3696 break;
3697 case MORIBUND:
3698 case ABORTING:
3699 case DEAD:
3700 pr_debug("ignoring disconnect ep %p state %u\n",
3701 ep, ep->com.state);
3702 break;
3703 default:
3704 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3705 break;
3708 if (close) {
3709 if (abrupt) {
3710 set_bit(EP_DISC_ABORT, &ep->com.history);
3711 ret = send_abort(ep);
3712 } else {
3713 set_bit(EP_DISC_CLOSE, &ep->com.history);
3714 ret = send_halfclose(ep);
3716 if (ret) {
3717 set_bit(EP_DISC_FAIL, &ep->com.history);
3718 if (!abrupt) {
3719 stop_ep_timer(ep);
3720 close_complete_upcall(ep, -EIO);
3722 if (ep->com.qp) {
3723 struct c4iw_qp_attributes attrs;
3725 attrs.next_state = C4IW_QP_STATE_ERROR;
3726 ret = c4iw_modify_qp(ep->com.qp->rhp,
3727 ep->com.qp,
3728 C4IW_QP_ATTR_NEXT_STATE,
3729 &attrs, 1);
3730 if (ret)
3731 pr_err("%s - qp <- error failed!\n",
3732 __func__);
3734 fatal = 1;
3737 mutex_unlock(&ep->com.mutex);
3738 c4iw_put_ep(&ep->com);
3739 if (fatal)
3740 release_ep_resources(ep);
3741 return ret;
3744 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3745 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3747 struct c4iw_ep *ep;
3748 int atid = be32_to_cpu(req->tid);
3750 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3751 (__force u32) req->tid);
3752 if (!ep)
3753 return;
3755 switch (req->retval) {
3756 case FW_ENOMEM:
3757 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3758 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3759 send_fw_act_open_req(ep, atid);
3760 return;
3762 fallthrough;
3763 case FW_EADDRINUSE:
3764 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3765 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3766 send_fw_act_open_req(ep, atid);
3767 return;
3769 break;
3770 default:
3771 pr_info("%s unexpected ofld conn wr retval %d\n",
3772 __func__, req->retval);
3773 break;
3775 pr_err("active ofld_connect_wr failure %d atid %d\n",
3776 req->retval, atid);
3777 mutex_lock(&dev->rdev.stats.lock);
3778 dev->rdev.stats.act_ofld_conn_fails++;
3779 mutex_unlock(&dev->rdev.stats.lock);
3780 connect_reply_upcall(ep, status2errno(req->retval));
3781 state_set(&ep->com, DEAD);
3782 if (ep->com.remote_addr.ss_family == AF_INET6) {
3783 struct sockaddr_in6 *sin6 =
3784 (struct sockaddr_in6 *)&ep->com.local_addr;
3785 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3786 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3788 xa_erase_irq(&dev->atids, atid);
3789 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3790 dst_release(ep->dst);
3791 cxgb4_l2t_release(ep->l2t);
3792 c4iw_put_ep(&ep->com);
3795 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3796 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3798 struct sk_buff *rpl_skb;
3799 struct cpl_pass_accept_req *cpl;
3800 int ret;
3802 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3803 if (req->retval) {
3804 pr_err("%s passive open failure %d\n", __func__, req->retval);
3805 mutex_lock(&dev->rdev.stats.lock);
3806 dev->rdev.stats.pas_ofld_conn_fails++;
3807 mutex_unlock(&dev->rdev.stats.lock);
3808 kfree_skb(rpl_skb);
3809 } else {
3810 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3811 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3812 (__force u32) htonl(
3813 (__force u32) req->tid)));
3814 ret = pass_accept_req(dev, rpl_skb);
3815 if (!ret)
3816 kfree_skb(rpl_skb);
3818 return;
3821 static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word)
3823 u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]);
3824 u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]);
3825 u64 t;
3826 u32 shift = 32;
3828 t = (thi << shift) | (tlo >> shift);
3830 return t;
3833 static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift)
3835 u32 v;
3836 u64 t = be64_to_cpu(tcb[(31 - word) / 2]);
3838 if (word & 0x1)
3839 shift += 32;
3840 v = (t >> shift) & mask;
3841 return v;
3844 static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3846 struct cpl_get_tcb_rpl *rpl = cplhdr(skb);
3847 __be64 *tcb = (__be64 *)(rpl + 1);
3848 unsigned int tid = GET_TID(rpl);
3849 struct c4iw_ep *ep;
3850 u64 t_flags_64;
3851 u32 rx_pdu_out;
3853 ep = get_ep_from_tid(dev, tid);
3854 if (!ep)
3855 return 0;
3856 /* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to
3857 * determine if there's a rx PDU feedback event pending.
3859 * If that bit is set, it means we'll need to re-read the TCB's
3860 * rq_start value. The final value is the one present in a TCB
3861 * with the TF_RX_PDU_OUT bit cleared.
3864 t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W);
3865 rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S;
3867 c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */
3868 c4iw_put_ep(&ep->com); /* from read_tcb() */
3870 /* If TF_RX_PDU_OUT bit is set, re-read the TCB */
3871 if (rx_pdu_out) {
3872 if (++ep->rx_pdu_out_cnt >= 2) {
3873 WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n");
3874 goto cleanup;
3876 read_tcb(ep);
3877 return 0;
3880 ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_M,
3881 TCB_RQ_START_S);
3882 cleanup:
3883 pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx);
3885 if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags))
3886 finish_peer_abort(dev, ep);
3887 else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags))
3888 send_abort_req(ep);
3889 else
3890 WARN_ONCE(1, "unexpected state!");
3892 return 0;
3895 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3897 struct cpl_fw6_msg *rpl = cplhdr(skb);
3898 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3900 switch (rpl->type) {
3901 case FW6_TYPE_CQE:
3902 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3903 break;
3904 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3905 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3906 switch (req->t_state) {
3907 case TCP_SYN_SENT:
3908 active_ofld_conn_reply(dev, skb, req);
3909 break;
3910 case TCP_SYN_RECV:
3911 passive_ofld_conn_reply(dev, skb, req);
3912 break;
3913 default:
3914 pr_err("%s unexpected ofld conn wr state %d\n",
3915 __func__, req->t_state);
3916 break;
3918 break;
3920 return 0;
3923 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3925 __be32 l2info;
3926 __be16 hdr_len, vlantag, len;
3927 u16 eth_hdr_len;
3928 int tcp_hdr_len, ip_hdr_len;
3929 u8 intf;
3930 struct cpl_rx_pkt *cpl = cplhdr(skb);
3931 struct cpl_pass_accept_req *req;
3932 struct tcp_options_received tmp_opt;
3933 struct c4iw_dev *dev;
3934 enum chip_type type;
3936 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3937 /* Store values from cpl_rx_pkt in temporary location. */
3938 vlantag = cpl->vlan;
3939 len = cpl->len;
3940 l2info = cpl->l2info;
3941 hdr_len = cpl->hdr_len;
3942 intf = cpl->iff;
3944 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3947 * We need to parse the TCP options from SYN packet.
3948 * to generate cpl_pass_accept_req.
3950 memset(&tmp_opt, 0, sizeof(tmp_opt));
3951 tcp_clear_options(&tmp_opt);
3952 tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
3954 req = __skb_push(skb, sizeof(*req));
3955 memset(req, 0, sizeof(*req));
3956 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3957 SYN_MAC_IDX_V(RX_MACIDX_G(
3958 be32_to_cpu(l2info))) |
3959 SYN_XACT_MATCH_F);
3960 type = dev->rdev.lldi.adapter_type;
3961 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3962 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3963 req->hdr_len =
3964 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3965 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3966 eth_hdr_len = is_t4(type) ?
3967 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3968 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3969 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3970 IP_HDR_LEN_V(ip_hdr_len) |
3971 ETH_HDR_LEN_V(eth_hdr_len));
3972 } else { /* T6 and later */
3973 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3974 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3975 T6_IP_HDR_LEN_V(ip_hdr_len) |
3976 T6_ETH_HDR_LEN_V(eth_hdr_len));
3978 req->vlan = vlantag;
3979 req->len = len;
3980 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3981 PASS_OPEN_TOS_V(tos));
3982 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3983 if (tmp_opt.wscale_ok)
3984 req->tcpopt.wsf = tmp_opt.snd_wscale;
3985 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3986 if (tmp_opt.sack_ok)
3987 req->tcpopt.sack = 1;
3988 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3989 return;
3992 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3993 __be32 laddr, __be16 lport,
3994 __be32 raddr, __be16 rport,
3995 u32 rcv_isn, u32 filter, u16 window,
3996 u32 rss_qid, u8 port_id)
3998 struct sk_buff *req_skb;
3999 struct fw_ofld_connection_wr *req;
4000 struct cpl_pass_accept_req *cpl = cplhdr(skb);
4001 int ret;
4003 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
4004 if (!req_skb)
4005 return;
4006 req = __skb_put_zero(req_skb, sizeof(*req));
4007 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
4008 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
4009 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
4010 req->le.filter = (__force __be32) filter;
4011 req->le.lport = lport;
4012 req->le.pport = rport;
4013 req->le.u.ipv4.lip = laddr;
4014 req->le.u.ipv4.pip = raddr;
4015 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
4016 req->tcb.rcv_adv = htons(window);
4017 req->tcb.t_state_to_astid =
4018 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
4019 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
4020 FW_OFLD_CONNECTION_WR_ASTID_V(
4021 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
4024 * We store the qid in opt2 which will be used by the firmware
4025 * to send us the wr response.
4027 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
4030 * We initialize the MSS index in TCB to 0xF.
4031 * So that when driver sends cpl_pass_accept_rpl
4032 * TCB picks up the correct value. If this was 0
4033 * TP will ignore any value > 0 for MSS index.
4035 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
4036 req->cookie = (uintptr_t)skb;
4038 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
4039 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
4040 if (ret < 0) {
4041 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
4042 ret);
4043 kfree_skb(skb);
4044 kfree_skb(req_skb);
4049 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
4050 * messages when a filter is being used instead of server to
4051 * redirect a syn packet. When packets hit filter they are redirected
4052 * to the offload queue and driver tries to establish the connection
4053 * using firmware work request.
4055 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
4057 int stid;
4058 unsigned int filter;
4059 struct ethhdr *eh = NULL;
4060 struct vlan_ethhdr *vlan_eh = NULL;
4061 struct iphdr *iph;
4062 struct tcphdr *tcph;
4063 struct rss_header *rss = (void *)skb->data;
4064 struct cpl_rx_pkt *cpl = (void *)skb->data;
4065 struct cpl_pass_accept_req *req = (void *)(rss + 1);
4066 struct l2t_entry *e;
4067 struct dst_entry *dst;
4068 struct c4iw_ep *lep = NULL;
4069 u16 window;
4070 struct port_info *pi;
4071 struct net_device *pdev;
4072 u16 rss_qid, eth_hdr_len;
4073 int step;
4074 struct neighbour *neigh;
4076 /* Drop all non-SYN packets */
4077 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
4078 goto reject;
4081 * Drop all packets which did not hit the filter.
4082 * Unlikely to happen.
4084 if (!(rss->filter_hit && rss->filter_tid))
4085 goto reject;
4088 * Calculate the server tid from filter hit index from cpl_rx_pkt.
4090 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4092 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4093 if (!lep) {
4094 pr_warn("%s connect request on invalid stid %d\n",
4095 __func__, stid);
4096 goto reject;
4099 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4100 case CHELSIO_T4:
4101 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4102 break;
4103 case CHELSIO_T5:
4104 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4105 break;
4106 case CHELSIO_T6:
4107 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4108 break;
4109 default:
4110 pr_err("T%d Chip is not supported\n",
4111 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4112 goto reject;
4115 if (eth_hdr_len == ETH_HLEN) {
4116 eh = (struct ethhdr *)(req + 1);
4117 iph = (struct iphdr *)(eh + 1);
4118 } else {
4119 vlan_eh = (struct vlan_ethhdr *)(req + 1);
4120 iph = (struct iphdr *)(vlan_eh + 1);
4121 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(cpl->vlan));
4124 if (iph->version != 0x4)
4125 goto reject;
4127 tcph = (struct tcphdr *)(iph + 1);
4128 skb_set_network_header(skb, (void *)iph - (void *)rss);
4129 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4130 skb_get(skb);
4132 pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n",
4133 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4134 ntohs(tcph->source), iph->tos);
4136 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
4137 iph->daddr, iph->saddr, tcph->dest,
4138 tcph->source, iph->tos);
4139 if (!dst) {
4140 pr_err("%s - failed to find dst entry!\n", __func__);
4141 goto reject;
4143 neigh = dst_neigh_lookup_skb(dst, skb);
4145 if (!neigh) {
4146 pr_err("%s - failed to allocate neigh!\n", __func__);
4147 goto free_dst;
4150 if (neigh->dev->flags & IFF_LOOPBACK) {
4151 pdev = ip_dev_find(&init_net, iph->daddr);
4152 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4153 pdev, 0);
4154 pi = (struct port_info *)netdev_priv(pdev);
4155 dev_put(pdev);
4156 } else {
4157 pdev = get_real_dev(neigh->dev);
4158 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4159 pdev, 0);
4160 pi = (struct port_info *)netdev_priv(pdev);
4162 neigh_release(neigh);
4163 if (!e) {
4164 pr_err("%s - failed to allocate l2t entry!\n",
4165 __func__);
4166 goto free_dst;
4169 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4170 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4171 window = (__force u16) htons((__force u16)tcph->window);
4173 /* Calcuate filter portion for LE region. */
4174 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4175 dev->rdev.lldi.ports[0],
4176 e));
4179 * Synthesize the cpl_pass_accept_req. We have everything except the
4180 * TID. Once firmware sends a reply with TID we update the TID field
4181 * in cpl and pass it through the regular cpl_pass_accept_req path.
4183 build_cpl_pass_accept_req(skb, stid, iph->tos);
4184 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4185 tcph->source, ntohl(tcph->seq), filter, window,
4186 rss_qid, pi->port_id);
4187 cxgb4_l2t_release(e);
4188 free_dst:
4189 dst_release(dst);
4190 reject:
4191 if (lep)
4192 c4iw_put_ep(&lep->com);
4193 return 0;
4197 * These are the real handlers that are called from a
4198 * work queue.
4200 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4201 [CPL_ACT_ESTABLISH] = act_establish,
4202 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4203 [CPL_RX_DATA] = rx_data,
4204 [CPL_ABORT_RPL_RSS] = abort_rpl,
4205 [CPL_ABORT_RPL] = abort_rpl,
4206 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4207 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4208 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4209 [CPL_PASS_ESTABLISH] = pass_establish,
4210 [CPL_PEER_CLOSE] = peer_close,
4211 [CPL_ABORT_REQ_RSS] = peer_abort,
4212 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4213 [CPL_RDMA_TERMINATE] = terminate,
4214 [CPL_FW4_ACK] = fw4_ack,
4215 [CPL_GET_TCB_RPL] = read_tcb_rpl,
4216 [CPL_FW6_MSG] = deferred_fw6_msg,
4217 [CPL_RX_PKT] = rx_pkt,
4218 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4219 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4222 static void process_timeout(struct c4iw_ep *ep)
4224 struct c4iw_qp_attributes attrs;
4225 int abort = 1;
4227 mutex_lock(&ep->com.mutex);
4228 pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state);
4229 set_bit(TIMEDOUT, &ep->com.history);
4230 switch (ep->com.state) {
4231 case MPA_REQ_SENT:
4232 connect_reply_upcall(ep, -ETIMEDOUT);
4233 break;
4234 case MPA_REQ_WAIT:
4235 case MPA_REQ_RCVD:
4236 case MPA_REP_SENT:
4237 case FPDU_MODE:
4238 break;
4239 case CLOSING:
4240 case MORIBUND:
4241 if (ep->com.cm_id && ep->com.qp) {
4242 attrs.next_state = C4IW_QP_STATE_ERROR;
4243 c4iw_modify_qp(ep->com.qp->rhp,
4244 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4245 &attrs, 1);
4247 close_complete_upcall(ep, -ETIMEDOUT);
4248 break;
4249 case ABORTING:
4250 case DEAD:
4253 * These states are expected if the ep timed out at the same
4254 * time as another thread was calling stop_ep_timer().
4255 * So we silently do nothing for these states.
4257 abort = 0;
4258 break;
4259 default:
4260 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4261 __func__, ep, ep->hwtid, ep->com.state);
4262 abort = 0;
4264 mutex_unlock(&ep->com.mutex);
4265 if (abort)
4266 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4267 c4iw_put_ep(&ep->com);
4270 static void process_timedout_eps(void)
4272 struct c4iw_ep *ep;
4274 spin_lock_irq(&timeout_lock);
4275 while (!list_empty(&timeout_list)) {
4276 struct list_head *tmp;
4278 tmp = timeout_list.next;
4279 list_del(tmp);
4280 tmp->next = NULL;
4281 tmp->prev = NULL;
4282 spin_unlock_irq(&timeout_lock);
4283 ep = list_entry(tmp, struct c4iw_ep, entry);
4284 process_timeout(ep);
4285 spin_lock_irq(&timeout_lock);
4287 spin_unlock_irq(&timeout_lock);
4290 static void process_work(struct work_struct *work)
4292 struct sk_buff *skb = NULL;
4293 struct c4iw_dev *dev;
4294 struct cpl_act_establish *rpl;
4295 unsigned int opcode;
4296 int ret;
4298 process_timedout_eps();
4299 while ((skb = skb_dequeue(&rxq))) {
4300 rpl = cplhdr(skb);
4301 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4302 opcode = rpl->ot.opcode;
4304 if (opcode >= ARRAY_SIZE(work_handlers) ||
4305 !work_handlers[opcode]) {
4306 pr_err("No handler for opcode 0x%x.\n", opcode);
4307 kfree_skb(skb);
4308 } else {
4309 ret = work_handlers[opcode](dev, skb);
4310 if (!ret)
4311 kfree_skb(skb);
4313 process_timedout_eps();
4317 static DECLARE_WORK(skb_work, process_work);
4319 static void ep_timeout(struct timer_list *t)
4321 struct c4iw_ep *ep = from_timer(ep, t, timer);
4322 int kickit = 0;
4324 spin_lock(&timeout_lock);
4325 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4327 * Only insert if it is not already on the list.
4329 if (!ep->entry.next) {
4330 list_add_tail(&ep->entry, &timeout_list);
4331 kickit = 1;
4334 spin_unlock(&timeout_lock);
4335 if (kickit)
4336 queue_work(workq, &skb_work);
4340 * All the CM events are handled on a work queue to have a safe context.
4342 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4346 * Save dev in the skb->cb area.
4348 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4351 * Queue the skb and schedule the worker thread.
4353 skb_queue_tail(&rxq, skb);
4354 queue_work(workq, &skb_work);
4355 return 0;
4358 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4360 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4362 if (rpl->status != CPL_ERR_NONE) {
4363 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4364 rpl->status, GET_TID(rpl));
4366 kfree_skb(skb);
4367 return 0;
4370 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4372 struct cpl_fw6_msg *rpl = cplhdr(skb);
4373 struct c4iw_wr_wait *wr_waitp;
4374 int ret;
4376 pr_debug("type %u\n", rpl->type);
4378 switch (rpl->type) {
4379 case FW6_TYPE_WR_RPL:
4380 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4381 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4382 pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret);
4383 if (wr_waitp)
4384 c4iw_wake_up_deref(wr_waitp, ret ? -ret : 0);
4385 kfree_skb(skb);
4386 break;
4387 case FW6_TYPE_CQE:
4388 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4389 sched(dev, skb);
4390 break;
4391 default:
4392 pr_err("%s unexpected fw6 msg type %u\n",
4393 __func__, rpl->type);
4394 kfree_skb(skb);
4395 break;
4397 return 0;
4400 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4402 struct cpl_abort_req_rss *req = cplhdr(skb);
4403 struct c4iw_ep *ep;
4404 unsigned int tid = GET_TID(req);
4406 ep = get_ep_from_tid(dev, tid);
4407 /* This EP will be dereferenced in peer_abort() */
4408 if (!ep) {
4409 pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4410 kfree_skb(skb);
4411 return 0;
4413 if (cxgb_is_neg_adv(req->status)) {
4414 pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
4415 ep->hwtid, req->status,
4416 neg_adv_str(req->status));
4417 goto out;
4419 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state);
4421 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
4422 out:
4423 sched(dev, skb);
4424 return 0;
4428 * Most upcalls from the T4 Core go to sched() to
4429 * schedule the processing on a work queue.
4431 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4432 [CPL_ACT_ESTABLISH] = sched,
4433 [CPL_ACT_OPEN_RPL] = sched,
4434 [CPL_RX_DATA] = sched,
4435 [CPL_ABORT_RPL_RSS] = sched,
4436 [CPL_ABORT_RPL] = sched,
4437 [CPL_PASS_OPEN_RPL] = sched,
4438 [CPL_CLOSE_LISTSRV_RPL] = sched,
4439 [CPL_PASS_ACCEPT_REQ] = sched,
4440 [CPL_PASS_ESTABLISH] = sched,
4441 [CPL_PEER_CLOSE] = sched,
4442 [CPL_CLOSE_CON_RPL] = sched,
4443 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4444 [CPL_RDMA_TERMINATE] = sched,
4445 [CPL_FW4_ACK] = sched,
4446 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4447 [CPL_GET_TCB_RPL] = sched,
4448 [CPL_FW6_MSG] = fw6_msg,
4449 [CPL_RX_PKT] = sched
4452 int __init c4iw_cm_init(void)
4454 spin_lock_init(&timeout_lock);
4455 skb_queue_head_init(&rxq);
4457 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4458 if (!workq)
4459 return -ENOMEM;
4461 return 0;
4464 void c4iw_cm_term(void)
4466 WARN_ON(!list_empty(&timeout_list));
4467 flush_workqueue(workq);
4468 destroy_workqueue(workq);