1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2016 Chelsio Communications, Inc.
6 #include <linux/module.h>
7 #include <linux/list.h>
8 #include <linux/workqueue.h>
9 #include <linux/skbuff.h>
10 #include <linux/timer.h>
11 #include <linux/notifier.h>
12 #include <linux/inetdevice.h>
14 #include <linux/tcp.h>
15 #include <linux/if_vlan.h>
17 #include <net/neighbour.h>
18 #include <net/netevent.h>
19 #include <net/route.h>
21 #include <net/ip6_route.h>
22 #include <net/addrconf.h>
24 #include <libcxgb_cm.h>
28 static void cxgbit_init_wr_wait(struct cxgbit_wr_wait *wr_waitp)
31 reinit_completion(&wr_waitp->completion);
35 cxgbit_wake_up(struct cxgbit_wr_wait *wr_waitp, const char *func, u8 ret)
37 if (ret == CPL_ERR_NONE)
43 pr_err("%s: err:%u", func, ret);
45 complete(&wr_waitp->completion);
49 cxgbit_wait_for_reply(struct cxgbit_device *cdev,
50 struct cxgbit_wr_wait *wr_waitp, u32 tid, u32 timeout,
55 if (!test_bit(CDEV_STATE_UP, &cdev->flags)) {
60 ret = wait_for_completion_timeout(&wr_waitp->completion, timeout * HZ);
62 pr_info("%s - Device %s not responding tid %u\n",
63 func, pci_name(cdev->lldi.pdev), tid);
64 wr_waitp->ret = -ETIMEDOUT;
68 pr_info("%s: FW reply %d tid %u\n",
69 pci_name(cdev->lldi.pdev), wr_waitp->ret, tid);
73 static int cxgbit_np_hashfn(const struct cxgbit_np *cnp)
75 return ((unsigned long)cnp >> 10) & (NP_INFO_HASH_SIZE - 1);
78 static struct np_info *
79 cxgbit_np_hash_add(struct cxgbit_device *cdev, struct cxgbit_np *cnp,
82 struct np_info *p = kzalloc(sizeof(*p), GFP_KERNEL);
85 int bucket = cxgbit_np_hashfn(cnp);
89 spin_lock(&cdev->np_lock);
90 p->next = cdev->np_hash_tab[bucket];
91 cdev->np_hash_tab[bucket] = p;
92 spin_unlock(&cdev->np_lock);
99 cxgbit_np_hash_find(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
101 int stid = -1, bucket = cxgbit_np_hashfn(cnp);
104 spin_lock(&cdev->np_lock);
105 for (p = cdev->np_hash_tab[bucket]; p; p = p->next) {
111 spin_unlock(&cdev->np_lock);
116 static int cxgbit_np_hash_del(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
118 int stid = -1, bucket = cxgbit_np_hashfn(cnp);
119 struct np_info *p, **prev = &cdev->np_hash_tab[bucket];
121 spin_lock(&cdev->np_lock);
122 for (p = *prev; p; prev = &p->next, p = p->next) {
130 spin_unlock(&cdev->np_lock);
135 void _cxgbit_free_cnp(struct kref *kref)
137 struct cxgbit_np *cnp;
139 cnp = container_of(kref, struct cxgbit_np, kref);
144 cxgbit_create_server6(struct cxgbit_device *cdev, unsigned int stid,
145 struct cxgbit_np *cnp)
147 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
148 &cnp->com.local_addr;
152 pr_debug("%s: dev = %s; stid = %u; sin6_port = %u\n",
153 __func__, cdev->lldi.ports[0]->name, stid, sin6->sin6_port);
155 addr_type = ipv6_addr_type((const struct in6_addr *)
157 if (addr_type != IPV6_ADDR_ANY) {
158 ret = cxgb4_clip_get(cdev->lldi.ports[0],
159 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
161 pr_err("Unable to find clip table entry. laddr %pI6. Error:%d.\n",
162 sin6->sin6_addr.s6_addr, ret);
168 cxgbit_init_wr_wait(&cnp->com.wr_wait);
170 ret = cxgb4_create_server6(cdev->lldi.ports[0],
171 stid, &sin6->sin6_addr,
173 cdev->lldi.rxq_ids[0]);
175 ret = cxgbit_wait_for_reply(cdev, &cnp->com.wr_wait,
178 ret = net_xmit_errno(ret);
183 if (ret != -ETIMEDOUT)
184 cxgb4_clip_release(cdev->lldi.ports[0],
185 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
187 pr_err("create server6 err %d stid %d laddr %pI6 lport %d\n",
188 ret, stid, sin6->sin6_addr.s6_addr,
189 ntohs(sin6->sin6_port));
196 cxgbit_create_server4(struct cxgbit_device *cdev, unsigned int stid,
197 struct cxgbit_np *cnp)
199 struct sockaddr_in *sin = (struct sockaddr_in *)
200 &cnp->com.local_addr;
203 pr_debug("%s: dev = %s; stid = %u; sin_port = %u\n",
204 __func__, cdev->lldi.ports[0]->name, stid, sin->sin_port);
207 cxgbit_init_wr_wait(&cnp->com.wr_wait);
209 ret = cxgb4_create_server(cdev->lldi.ports[0],
210 stid, sin->sin_addr.s_addr,
212 cdev->lldi.rxq_ids[0]);
214 ret = cxgbit_wait_for_reply(cdev,
218 ret = net_xmit_errno(ret);
223 pr_err("create server failed err %d stid %d laddr %pI4 lport %d\n",
224 ret, stid, &sin->sin_addr, ntohs(sin->sin_port));
228 struct cxgbit_device *cxgbit_find_device(struct net_device *ndev, u8 *port_id)
230 struct cxgbit_device *cdev;
233 list_for_each_entry(cdev, &cdev_list_head, list) {
234 struct cxgb4_lld_info *lldi = &cdev->lldi;
236 for (i = 0; i < lldi->nports; i++) {
237 if (lldi->ports[i] == ndev) {
248 static struct net_device *cxgbit_get_real_dev(struct net_device *ndev)
250 if (ndev->priv_flags & IFF_BONDING) {
251 pr_err("Bond devices are not supported. Interface:%s\n",
256 if (is_vlan_dev(ndev))
257 return vlan_dev_real_dev(ndev);
262 static struct net_device *cxgbit_ipv4_netdev(__be32 saddr)
264 struct net_device *ndev;
266 ndev = __ip_dev_find(&init_net, saddr, false);
270 return cxgbit_get_real_dev(ndev);
273 static struct net_device *cxgbit_ipv6_netdev(struct in6_addr *addr6)
275 struct net_device *ndev = NULL;
278 if (IS_ENABLED(CONFIG_IPV6)) {
279 for_each_netdev_rcu(&init_net, ndev)
280 if (ipv6_chk_addr(&init_net, addr6, ndev, 1)) {
287 return cxgbit_get_real_dev(ndev);
290 static struct cxgbit_device *cxgbit_find_np_cdev(struct cxgbit_np *cnp)
292 struct sockaddr_storage *sockaddr = &cnp->com.local_addr;
293 int ss_family = sockaddr->ss_family;
294 struct net_device *ndev = NULL;
295 struct cxgbit_device *cdev = NULL;
298 if (ss_family == AF_INET) {
299 struct sockaddr_in *sin;
301 sin = (struct sockaddr_in *)sockaddr;
302 ndev = cxgbit_ipv4_netdev(sin->sin_addr.s_addr);
303 } else if (ss_family == AF_INET6) {
304 struct sockaddr_in6 *sin6;
306 sin6 = (struct sockaddr_in6 *)sockaddr;
307 ndev = cxgbit_ipv6_netdev(&sin6->sin6_addr);
312 cdev = cxgbit_find_device(ndev, NULL);
318 static bool cxgbit_inaddr_any(struct cxgbit_np *cnp)
320 struct sockaddr_storage *sockaddr = &cnp->com.local_addr;
321 int ss_family = sockaddr->ss_family;
324 if (ss_family == AF_INET) {
325 struct sockaddr_in *sin;
327 sin = (struct sockaddr_in *)sockaddr;
328 if (sin->sin_addr.s_addr == htonl(INADDR_ANY))
330 } else if (ss_family == AF_INET6) {
331 struct sockaddr_in6 *sin6;
333 sin6 = (struct sockaddr_in6 *)sockaddr;
334 addr_type = ipv6_addr_type((const struct in6_addr *)
336 if (addr_type == IPV6_ADDR_ANY)
343 __cxgbit_setup_cdev_np(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
346 int ss_family = cnp->com.local_addr.ss_family;
348 if (!test_bit(CDEV_STATE_UP, &cdev->flags))
351 stid = cxgb4_alloc_stid(cdev->lldi.tids, ss_family, cnp);
355 if (!cxgbit_np_hash_add(cdev, cnp, stid)) {
356 cxgb4_free_stid(cdev->lldi.tids, stid, ss_family);
360 if (ss_family == AF_INET)
361 ret = cxgbit_create_server4(cdev, stid, cnp);
363 ret = cxgbit_create_server6(cdev, stid, cnp);
366 if (ret != -ETIMEDOUT)
367 cxgb4_free_stid(cdev->lldi.tids, stid,
369 cxgbit_np_hash_del(cdev, cnp);
375 static int cxgbit_setup_cdev_np(struct cxgbit_np *cnp)
377 struct cxgbit_device *cdev;
380 mutex_lock(&cdev_list_lock);
381 cdev = cxgbit_find_np_cdev(cnp);
385 if (cxgbit_np_hash_find(cdev, cnp) >= 0)
388 if (__cxgbit_setup_cdev_np(cdev, cnp))
391 cnp->com.cdev = cdev;
394 mutex_unlock(&cdev_list_lock);
398 static int cxgbit_setup_all_np(struct cxgbit_np *cnp)
400 struct cxgbit_device *cdev;
404 mutex_lock(&cdev_list_lock);
405 list_for_each_entry(cdev, &cdev_list_head, list) {
406 if (cxgbit_np_hash_find(cdev, cnp) >= 0) {
407 mutex_unlock(&cdev_list_lock);
412 list_for_each_entry(cdev, &cdev_list_head, list) {
413 ret = __cxgbit_setup_cdev_np(cdev, cnp);
414 if (ret == -ETIMEDOUT)
420 mutex_unlock(&cdev_list_lock);
422 return count ? 0 : -1;
425 int cxgbit_setup_np(struct iscsi_np *np, struct sockaddr_storage *ksockaddr)
427 struct cxgbit_np *cnp;
430 if ((ksockaddr->ss_family != AF_INET) &&
431 (ksockaddr->ss_family != AF_INET6))
434 cnp = kzalloc(sizeof(*cnp), GFP_KERNEL);
438 init_waitqueue_head(&cnp->accept_wait);
439 init_completion(&cnp->com.wr_wait.completion);
440 init_completion(&cnp->accept_comp);
441 INIT_LIST_HEAD(&cnp->np_accept_list);
442 spin_lock_init(&cnp->np_accept_lock);
443 kref_init(&cnp->kref);
444 memcpy(&np->np_sockaddr, ksockaddr,
445 sizeof(struct sockaddr_storage));
446 memcpy(&cnp->com.local_addr, &np->np_sockaddr,
447 sizeof(cnp->com.local_addr));
450 cnp->com.cdev = NULL;
452 if (cxgbit_inaddr_any(cnp))
453 ret = cxgbit_setup_all_np(cnp);
455 ret = cxgbit_setup_cdev_np(cnp);
462 np->np_context = cnp;
463 cnp->com.state = CSK_STATE_LISTEN;
468 cxgbit_set_conn_info(struct iscsi_np *np, struct iscsi_conn *conn,
469 struct cxgbit_sock *csk)
471 conn->login_family = np->np_sockaddr.ss_family;
472 conn->login_sockaddr = csk->com.remote_addr;
473 conn->local_sockaddr = csk->com.local_addr;
476 int cxgbit_accept_np(struct iscsi_np *np, struct iscsi_conn *conn)
478 struct cxgbit_np *cnp = np->np_context;
479 struct cxgbit_sock *csk;
483 ret = wait_for_completion_interruptible(&cnp->accept_comp);
487 spin_lock_bh(&np->np_thread_lock);
488 if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) {
489 spin_unlock_bh(&np->np_thread_lock);
491 * No point in stalling here when np_thread
492 * is in state RESET/SHUTDOWN/EXIT - bail
496 spin_unlock_bh(&np->np_thread_lock);
498 spin_lock_bh(&cnp->np_accept_lock);
499 if (list_empty(&cnp->np_accept_list)) {
500 spin_unlock_bh(&cnp->np_accept_lock);
504 csk = list_first_entry(&cnp->np_accept_list,
508 list_del_init(&csk->accept_node);
509 spin_unlock_bh(&cnp->np_accept_lock);
513 cxgbit_set_conn_info(np, conn, csk);
518 __cxgbit_free_cdev_np(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
523 stid = cxgbit_np_hash_del(cdev, cnp);
526 if (!test_bit(CDEV_STATE_UP, &cdev->flags))
529 if (cnp->np->np_sockaddr.ss_family == AF_INET6)
533 cxgbit_init_wr_wait(&cnp->com.wr_wait);
534 ret = cxgb4_remove_server(cdev->lldi.ports[0], stid,
535 cdev->lldi.rxq_ids[0], ipv6);
538 ret = net_xmit_errno(ret);
545 ret = cxgbit_wait_for_reply(cdev, &cnp->com.wr_wait,
547 if (ret == -ETIMEDOUT)
550 if (ipv6 && cnp->com.cdev) {
551 struct sockaddr_in6 *sin6;
553 sin6 = (struct sockaddr_in6 *)&cnp->com.local_addr;
554 cxgb4_clip_release(cdev->lldi.ports[0],
555 (const u32 *)&sin6->sin6_addr.s6_addr,
559 cxgb4_free_stid(cdev->lldi.tids, stid,
560 cnp->com.local_addr.ss_family);
564 static void cxgbit_free_all_np(struct cxgbit_np *cnp)
566 struct cxgbit_device *cdev;
569 mutex_lock(&cdev_list_lock);
570 list_for_each_entry(cdev, &cdev_list_head, list) {
571 ret = __cxgbit_free_cdev_np(cdev, cnp);
572 if (ret == -ETIMEDOUT)
575 mutex_unlock(&cdev_list_lock);
578 static void cxgbit_free_cdev_np(struct cxgbit_np *cnp)
580 struct cxgbit_device *cdev;
583 mutex_lock(&cdev_list_lock);
584 list_for_each_entry(cdev, &cdev_list_head, list) {
585 if (cdev == cnp->com.cdev) {
593 __cxgbit_free_cdev_np(cdev, cnp);
595 mutex_unlock(&cdev_list_lock);
598 static void __cxgbit_free_conn(struct cxgbit_sock *csk);
600 void cxgbit_free_np(struct iscsi_np *np)
602 struct cxgbit_np *cnp = np->np_context;
603 struct cxgbit_sock *csk, *tmp;
605 cnp->com.state = CSK_STATE_DEAD;
607 cxgbit_free_cdev_np(cnp);
609 cxgbit_free_all_np(cnp);
611 spin_lock_bh(&cnp->np_accept_lock);
612 list_for_each_entry_safe(csk, tmp, &cnp->np_accept_list, accept_node) {
613 list_del_init(&csk->accept_node);
614 __cxgbit_free_conn(csk);
616 spin_unlock_bh(&cnp->np_accept_lock);
618 np->np_context = NULL;
622 static void cxgbit_send_halfclose(struct cxgbit_sock *csk)
625 u32 len = roundup(sizeof(struct cpl_close_con_req), 16);
627 skb = alloc_skb(len, GFP_ATOMIC);
631 cxgb_mk_close_con_req(skb, len, csk->tid, csk->txq_idx,
634 cxgbit_skcb_flags(skb) |= SKCBF_TX_FLAG_COMPL;
635 __skb_queue_tail(&csk->txq, skb);
636 cxgbit_push_tx_frames(csk);
639 static void cxgbit_arp_failure_discard(void *handle, struct sk_buff *skb)
641 struct cxgbit_sock *csk = handle;
643 pr_debug("%s cxgbit_device %p\n", __func__, handle);
648 static void cxgbit_abort_arp_failure(void *handle, struct sk_buff *skb)
650 struct cxgbit_device *cdev = handle;
651 struct cpl_abort_req *req = cplhdr(skb);
653 pr_debug("%s cdev %p\n", __func__, cdev);
654 req->cmd = CPL_ABORT_NO_RST;
655 cxgbit_ofld_send(cdev, skb);
658 static int cxgbit_send_abort_req(struct cxgbit_sock *csk)
661 u32 len = roundup(sizeof(struct cpl_abort_req), 16);
663 pr_debug("%s: csk %p tid %u; state %d\n",
664 __func__, csk, csk->tid, csk->com.state);
666 __skb_queue_purge(&csk->txq);
668 if (!test_and_set_bit(CSK_TX_DATA_SENT, &csk->com.flags))
669 cxgbit_send_tx_flowc_wr(csk);
671 skb = __skb_dequeue(&csk->skbq);
672 cxgb_mk_abort_req(skb, len, csk->tid, csk->txq_idx,
673 csk->com.cdev, cxgbit_abort_arp_failure);
675 return cxgbit_l2t_send(csk->com.cdev, skb, csk->l2t);
679 __cxgbit_abort_conn(struct cxgbit_sock *csk, struct sk_buff *skb)
683 if (csk->com.state != CSK_STATE_ESTABLISHED)
686 set_bit(CSK_ABORT_RPL_WAIT, &csk->com.flags);
687 csk->com.state = CSK_STATE_ABORTING;
689 cxgbit_send_abort_req(csk);
694 cxgbit_wake_up(&csk->com.wr_wait, __func__, CPL_ERR_NONE);
698 void cxgbit_abort_conn(struct cxgbit_sock *csk)
700 struct sk_buff *skb = alloc_skb(0, GFP_KERNEL | __GFP_NOFAIL);
703 cxgbit_init_wr_wait(&csk->com.wr_wait);
705 spin_lock_bh(&csk->lock);
706 if (csk->lock_owner) {
707 cxgbit_skcb_rx_backlog_fn(skb) = __cxgbit_abort_conn;
708 __skb_queue_tail(&csk->backlogq, skb);
710 __cxgbit_abort_conn(csk, skb);
712 spin_unlock_bh(&csk->lock);
714 cxgbit_wait_for_reply(csk->com.cdev, &csk->com.wr_wait,
715 csk->tid, 600, __func__);
718 static void __cxgbit_free_conn(struct cxgbit_sock *csk)
720 struct iscsi_conn *conn = csk->conn;
721 bool release = false;
723 pr_debug("%s: state %d\n",
724 __func__, csk->com.state);
726 spin_lock_bh(&csk->lock);
727 switch (csk->com.state) {
728 case CSK_STATE_ESTABLISHED:
729 if (conn && (conn->conn_state == TARG_CONN_STATE_IN_LOGOUT)) {
730 csk->com.state = CSK_STATE_CLOSING;
731 cxgbit_send_halfclose(csk);
733 csk->com.state = CSK_STATE_ABORTING;
734 cxgbit_send_abort_req(csk);
737 case CSK_STATE_CLOSING:
738 csk->com.state = CSK_STATE_MORIBUND;
739 cxgbit_send_halfclose(csk);
745 pr_err("%s: csk %p; state %d\n",
746 __func__, csk, csk->com.state);
748 spin_unlock_bh(&csk->lock);
754 void cxgbit_free_conn(struct iscsi_conn *conn)
756 __cxgbit_free_conn(conn->context);
759 static void cxgbit_set_emss(struct cxgbit_sock *csk, u16 opt)
761 csk->emss = csk->com.cdev->lldi.mtus[TCPOPT_MSS_G(opt)] -
762 ((csk->com.remote_addr.ss_family == AF_INET) ?
763 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
764 sizeof(struct tcphdr);
765 csk->mss = csk->emss;
766 if (TCPOPT_TSTAMP_G(opt))
767 csk->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
771 pr_info("Warning: misaligned mtu idx %u mss %u emss=%u\n",
772 TCPOPT_MSS_G(opt), csk->mss, csk->emss);
773 pr_debug("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
774 csk->mss, csk->emss);
777 static void cxgbit_free_skb(struct cxgbit_sock *csk)
781 __skb_queue_purge(&csk->txq);
782 __skb_queue_purge(&csk->rxq);
783 __skb_queue_purge(&csk->backlogq);
784 __skb_queue_purge(&csk->ppodq);
785 __skb_queue_purge(&csk->skbq);
787 while ((skb = cxgbit_sock_dequeue_wr(csk)))
790 __kfree_skb(csk->lro_hskb);
793 void _cxgbit_free_csk(struct kref *kref)
795 struct cxgbit_sock *csk;
796 struct cxgbit_device *cdev;
798 csk = container_of(kref, struct cxgbit_sock, kref);
800 pr_debug("%s csk %p state %d\n", __func__, csk, csk->com.state);
802 if (csk->com.local_addr.ss_family == AF_INET6) {
803 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
804 &csk->com.local_addr;
805 cxgb4_clip_release(csk->com.cdev->lldi.ports[0],
807 &sin6->sin6_addr.s6_addr, 1);
810 cxgb4_remove_tid(csk->com.cdev->lldi.tids, 0, csk->tid,
811 csk->com.local_addr.ss_family);
812 dst_release(csk->dst);
813 cxgb4_l2t_release(csk->l2t);
815 cdev = csk->com.cdev;
816 spin_lock_bh(&cdev->cskq.lock);
817 list_del(&csk->list);
818 spin_unlock_bh(&cdev->cskq.lock);
820 cxgbit_free_skb(csk);
821 cxgbit_put_cnp(csk->cnp);
822 cxgbit_put_cdev(cdev);
827 static void cxgbit_set_tcp_window(struct cxgbit_sock *csk, struct port_info *pi)
829 unsigned int linkspeed;
832 linkspeed = pi->link_cfg.speed;
833 scale = linkspeed / SPEED_10000;
835 #define CXGBIT_10G_RCV_WIN (256 * 1024)
836 csk->rcv_win = CXGBIT_10G_RCV_WIN;
838 csk->rcv_win *= scale;
840 #define CXGBIT_10G_SND_WIN (256 * 1024)
841 csk->snd_win = CXGBIT_10G_SND_WIN;
843 csk->snd_win *= scale;
845 pr_debug("%s snd_win %d rcv_win %d\n",
846 __func__, csk->snd_win, csk->rcv_win);
849 #ifdef CONFIG_CHELSIO_T4_DCB
850 static u8 cxgbit_get_iscsi_dcb_state(struct net_device *ndev)
852 return ndev->dcbnl_ops->getstate(ndev);
855 static int cxgbit_select_priority(int pri_mask)
860 return (ffs(pri_mask) - 1);
863 static u8 cxgbit_get_iscsi_dcb_priority(struct net_device *ndev, u16 local_port)
868 struct dcb_app iscsi_dcb_app = {
869 .protocol = local_port
872 ret = (int)ndev->dcbnl_ops->getcap(ndev, DCB_CAP_ATTR_DCBX, &caps);
877 if (caps & DCB_CAP_DCBX_VER_IEEE) {
878 iscsi_dcb_app.selector = IEEE_8021QAZ_APP_SEL_STREAM;
879 ret = dcb_ieee_getapp_mask(ndev, &iscsi_dcb_app);
881 iscsi_dcb_app.selector = IEEE_8021QAZ_APP_SEL_ANY;
882 ret = dcb_ieee_getapp_mask(ndev, &iscsi_dcb_app);
884 } else if (caps & DCB_CAP_DCBX_VER_CEE) {
885 iscsi_dcb_app.selector = DCB_APP_IDTYPE_PORTNUM;
887 ret = dcb_getapp(ndev, &iscsi_dcb_app);
890 pr_info("iSCSI priority is set to %u\n", cxgbit_select_priority(ret));
892 return cxgbit_select_priority(ret);
897 cxgbit_offload_init(struct cxgbit_sock *csk, int iptype, __u8 *peer_ip,
898 u16 local_port, struct dst_entry *dst,
899 struct cxgbit_device *cdev)
903 struct net_device *ndev;
904 u16 rxq_idx, port_id;
905 #ifdef CONFIG_CHELSIO_T4_DCB
909 n = dst_neigh_lookup(dst, peer_ip);
914 if (!(n->nud_state & NUD_VALID))
915 neigh_event_send(n, NULL);
918 if (n->dev->flags & IFF_LOOPBACK) {
920 ndev = cxgbit_ipv4_netdev(*(__be32 *)peer_ip);
921 else if (IS_ENABLED(CONFIG_IPV6))
922 ndev = cxgbit_ipv6_netdev((struct in6_addr *)peer_ip);
931 csk->l2t = cxgb4_l2t_get(cdev->lldi.l2t,
935 csk->mtu = ndev->mtu;
936 csk->tx_chan = cxgb4_port_chan(ndev);
938 ((struct port_info *)netdev_priv(ndev))->smt_idx;
939 step = cdev->lldi.ntxq /
941 csk->txq_idx = cxgb4_port_idx(ndev) * step;
942 step = cdev->lldi.nrxq /
944 csk->ctrlq_idx = cxgb4_port_idx(ndev);
945 csk->rss_qid = cdev->lldi.rxq_ids[
946 cxgb4_port_idx(ndev) * step];
947 csk->port_id = cxgb4_port_idx(ndev);
948 cxgbit_set_tcp_window(csk,
949 (struct port_info *)netdev_priv(ndev));
951 ndev = cxgbit_get_real_dev(n->dev);
957 #ifdef CONFIG_CHELSIO_T4_DCB
958 if (cxgbit_get_iscsi_dcb_state(ndev))
959 priority = cxgbit_get_iscsi_dcb_priority(ndev,
962 csk->dcb_priority = priority;
964 csk->l2t = cxgb4_l2t_get(cdev->lldi.l2t, n, ndev, priority);
966 csk->l2t = cxgb4_l2t_get(cdev->lldi.l2t, n, ndev, 0);
970 port_id = cxgb4_port_idx(ndev);
971 csk->mtu = dst_mtu(dst);
972 csk->tx_chan = cxgb4_port_chan(ndev);
974 ((struct port_info *)netdev_priv(ndev))->smt_idx;
975 step = cdev->lldi.ntxq /
977 csk->txq_idx = (port_id * step) +
978 (cdev->selectq[port_id][0]++ % step);
979 csk->ctrlq_idx = cxgb4_port_idx(ndev);
980 step = cdev->lldi.nrxq /
982 rxq_idx = (port_id * step) +
983 (cdev->selectq[port_id][1]++ % step);
984 csk->rss_qid = cdev->lldi.rxq_ids[rxq_idx];
985 csk->port_id = port_id;
986 cxgbit_set_tcp_window(csk,
987 (struct port_info *)netdev_priv(ndev));
996 int cxgbit_ofld_send(struct cxgbit_device *cdev, struct sk_buff *skb)
1000 if (!test_bit(CDEV_STATE_UP, &cdev->flags)) {
1002 pr_err("%s - device not up - dropping\n", __func__);
1006 ret = cxgb4_ofld_send(cdev->lldi.ports[0], skb);
1009 return ret < 0 ? ret : 0;
1012 static void cxgbit_release_tid(struct cxgbit_device *cdev, u32 tid)
1014 u32 len = roundup(sizeof(struct cpl_tid_release), 16);
1015 struct sk_buff *skb;
1017 skb = alloc_skb(len, GFP_ATOMIC);
1021 cxgb_mk_tid_release(skb, len, tid, 0);
1022 cxgbit_ofld_send(cdev, skb);
1026 cxgbit_l2t_send(struct cxgbit_device *cdev, struct sk_buff *skb,
1027 struct l2t_entry *l2e)
1031 if (!test_bit(CDEV_STATE_UP, &cdev->flags)) {
1033 pr_err("%s - device not up - dropping\n", __func__);
1037 ret = cxgb4_l2t_send(cdev->lldi.ports[0], skb, l2e);
1040 return ret < 0 ? ret : 0;
1043 static void cxgbit_send_rx_credits(struct cxgbit_sock *csk, struct sk_buff *skb)
1045 if (csk->com.state != CSK_STATE_ESTABLISHED) {
1050 cxgbit_ofld_send(csk->com.cdev, skb);
1054 * CPL connection rx data ack: host ->
1055 * Send RX credits through an RX_DATA_ACK CPL message.
1056 * Returns the number of credits sent.
1058 int cxgbit_rx_data_ack(struct cxgbit_sock *csk)
1060 struct sk_buff *skb;
1061 u32 len = roundup(sizeof(struct cpl_rx_data_ack), 16);
1064 skb = alloc_skb(len, GFP_KERNEL);
1068 credit_dack = RX_DACK_CHANGE_F | RX_DACK_MODE_V(1) |
1069 RX_CREDITS_V(csk->rx_credits);
1071 cxgb_mk_rx_data_ack(skb, len, csk->tid, csk->ctrlq_idx,
1074 csk->rx_credits = 0;
1076 spin_lock_bh(&csk->lock);
1077 if (csk->lock_owner) {
1078 cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_send_rx_credits;
1079 __skb_queue_tail(&csk->backlogq, skb);
1080 spin_unlock_bh(&csk->lock);
1084 cxgbit_send_rx_credits(csk, skb);
1085 spin_unlock_bh(&csk->lock);
1090 #define FLOWC_WR_NPARAMS_MIN 9
1091 #define FLOWC_WR_NPARAMS_MAX 11
1092 static int cxgbit_alloc_csk_skb(struct cxgbit_sock *csk)
1094 struct sk_buff *skb;
1098 flowclen = offsetof(struct fw_flowc_wr,
1099 mnemval[FLOWC_WR_NPARAMS_MAX]);
1101 len = max_t(u32, sizeof(struct cpl_abort_req),
1102 sizeof(struct cpl_abort_rpl));
1104 len = max(len, flowclen);
1105 len = roundup(len, 16);
1107 for (i = 0; i < 3; i++) {
1108 skb = alloc_skb(len, GFP_ATOMIC);
1111 __skb_queue_tail(&csk->skbq, skb);
1114 skb = alloc_skb(LRO_SKB_MIN_HEADROOM, GFP_ATOMIC);
1118 memset(skb->data, 0, LRO_SKB_MIN_HEADROOM);
1119 csk->lro_hskb = skb;
1123 __skb_queue_purge(&csk->skbq);
1128 cxgbit_pass_accept_rpl(struct cxgbit_sock *csk, struct cpl_pass_accept_req *req)
1130 struct sk_buff *skb;
1131 const struct tcphdr *tcph;
1132 struct cpl_t5_pass_accept_rpl *rpl5;
1133 struct cxgb4_lld_info *lldi = &csk->com.cdev->lldi;
1134 unsigned int len = roundup(sizeof(*rpl5), 16);
1135 unsigned int mtu_idx;
1141 pr_debug("%s csk %p tid %u\n", __func__, csk, csk->tid);
1143 skb = alloc_skb(len, GFP_ATOMIC);
1145 cxgbit_put_csk(csk);
1149 rpl5 = __skb_put_zero(skb, len);
1151 INIT_TP_WR(rpl5, csk->tid);
1152 OPCODE_TID(rpl5) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1154 cxgb_best_mtu(csk->com.cdev->lldi.mtus, csk->mtu, &mtu_idx,
1156 (csk->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1157 wscale = cxgb_compute_wscale(csk->rcv_win);
1159 * Specify the largest window that will fit in opt0. The
1160 * remainder will be specified in the rx_data_ack.
1162 win = csk->rcv_win >> 10;
1163 if (win > RCV_BUFSIZ_M)
1165 opt0 = TCAM_BYPASS_F |
1166 WND_SCALE_V(wscale) |
1167 MSS_IDX_V(mtu_idx) |
1168 L2T_IDX_V(csk->l2t->idx) |
1169 TX_CHAN_V(csk->tx_chan) |
1170 SMAC_SEL_V(csk->smac_idx) |
1171 DSCP_V(csk->tos >> 2) |
1172 ULP_MODE_V(ULP_MODE_ISCSI) |
1175 opt2 = RX_CHANNEL_V(0) |
1176 RSS_QUEUE_VALID_F | RSS_QUEUE_V(csk->rss_qid);
1178 if (!is_t5(lldi->adapter_type))
1179 opt2 |= RX_FC_DISABLE_F;
1181 if (req->tcpopt.tstamp)
1182 opt2 |= TSTAMPS_EN_F;
1183 if (req->tcpopt.sack)
1186 opt2 |= WND_SCALE_EN_F;
1188 hlen = ntohl(req->hdr_len);
1190 if (is_t5(lldi->adapter_type))
1191 tcph = (struct tcphdr *)((u8 *)(req + 1) +
1192 ETH_HDR_LEN_G(hlen) + IP_HDR_LEN_G(hlen));
1194 tcph = (struct tcphdr *)((u8 *)(req + 1) +
1195 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen));
1197 if (tcph->ece && tcph->cwr)
1198 opt2 |= CCTRL_ECN_V(1);
1200 opt2 |= RX_COALESCE_V(3);
1201 opt2 |= CONG_CNTRL_V(CONG_ALG_NEWRENO);
1204 rpl5->iss = cpu_to_be32((prandom_u32() & ~7UL) - 1);
1206 opt2 |= T5_OPT_2_VALID_F;
1208 rpl5->opt0 = cpu_to_be64(opt0);
1209 rpl5->opt2 = cpu_to_be32(opt2);
1210 set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->ctrlq_idx);
1211 t4_set_arp_err_handler(skb, csk, cxgbit_arp_failure_discard);
1212 cxgbit_l2t_send(csk->com.cdev, skb, csk->l2t);
1216 cxgbit_pass_accept_req(struct cxgbit_device *cdev, struct sk_buff *skb)
1218 struct cxgbit_sock *csk = NULL;
1219 struct cxgbit_np *cnp;
1220 struct cpl_pass_accept_req *req = cplhdr(skb);
1221 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1222 struct tid_info *t = cdev->lldi.tids;
1223 unsigned int tid = GET_TID(req);
1224 u16 peer_mss = ntohs(req->tcpopt.mss);
1225 unsigned short hdrs;
1227 struct dst_entry *dst;
1228 __u8 local_ip[16], peer_ip[16];
1229 __be16 local_port, peer_port;
1233 pr_debug("%s: cdev = %p; stid = %u; tid = %u\n",
1234 __func__, cdev, stid, tid);
1236 cnp = lookup_stid(t, stid);
1238 pr_err("%s connect request on invalid stid %d\n",
1243 if (cnp->com.state != CSK_STATE_LISTEN) {
1244 pr_err("%s - listening parent not in CSK_STATE_LISTEN\n",
1249 csk = lookup_tid(t, tid);
1251 pr_err("%s csk not null tid %u\n",
1256 cxgb_get_4tuple(req, cdev->lldi.adapter_type, &iptype, local_ip,
1257 peer_ip, &local_port, &peer_port);
1259 /* Find output route */
1261 pr_debug("%s parent sock %p tid %u laddr %pI4 raddr %pI4 "
1262 "lport %d rport %d peer_mss %d\n"
1263 , __func__, cnp, tid,
1264 local_ip, peer_ip, ntohs(local_port),
1265 ntohs(peer_port), peer_mss);
1266 dst = cxgb_find_route(&cdev->lldi, cxgbit_get_real_dev,
1267 *(__be32 *)local_ip,
1269 local_port, peer_port,
1270 PASS_OPEN_TOS_G(ntohl(req->tos_stid)));
1272 pr_debug("%s parent sock %p tid %u laddr %pI6 raddr %pI6 "
1273 "lport %d rport %d peer_mss %d\n"
1274 , __func__, cnp, tid,
1275 local_ip, peer_ip, ntohs(local_port),
1276 ntohs(peer_port), peer_mss);
1277 dst = cxgb_find_route6(&cdev->lldi, cxgbit_get_real_dev,
1279 local_port, peer_port,
1280 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
1281 ((struct sockaddr_in6 *)
1282 &cnp->com.local_addr)->sin6_scope_id);
1285 pr_err("%s - failed to find dst entry!\n",
1290 csk = kzalloc(sizeof(*csk), GFP_ATOMIC);
1296 ret = cxgbit_offload_init(csk, iptype, peer_ip, ntohs(local_port),
1299 pr_err("%s - failed to allocate l2t entry!\n",
1306 kref_init(&csk->kref);
1307 init_completion(&csk->com.wr_wait.completion);
1309 INIT_LIST_HEAD(&csk->accept_node);
1311 hdrs = (iptype == 4 ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
1312 sizeof(struct tcphdr) + (req->tcpopt.tstamp ? 12 : 0);
1313 if (peer_mss && csk->mtu > (peer_mss + hdrs))
1314 csk->mtu = peer_mss + hdrs;
1316 csk->com.state = CSK_STATE_CONNECTING;
1317 csk->com.cdev = cdev;
1319 csk->tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
1322 csk->wr_cred = cdev->lldi.wr_cred -
1323 DIV_ROUND_UP(sizeof(struct cpl_abort_req), 16);
1324 csk->wr_max_cred = csk->wr_cred;
1325 csk->wr_una_cred = 0;
1328 struct sockaddr_in *sin = (struct sockaddr_in *)
1329 &csk->com.local_addr;
1330 sin->sin_family = AF_INET;
1331 sin->sin_port = local_port;
1332 sin->sin_addr.s_addr = *(__be32 *)local_ip;
1334 sin = (struct sockaddr_in *)&csk->com.remote_addr;
1335 sin->sin_family = AF_INET;
1336 sin->sin_port = peer_port;
1337 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
1339 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
1340 &csk->com.local_addr;
1342 sin6->sin6_family = PF_INET6;
1343 sin6->sin6_port = local_port;
1344 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
1345 cxgb4_clip_get(cdev->lldi.ports[0],
1346 (const u32 *)&sin6->sin6_addr.s6_addr,
1349 sin6 = (struct sockaddr_in6 *)&csk->com.remote_addr;
1350 sin6->sin6_family = PF_INET6;
1351 sin6->sin6_port = peer_port;
1352 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
1355 skb_queue_head_init(&csk->rxq);
1356 skb_queue_head_init(&csk->txq);
1357 skb_queue_head_init(&csk->ppodq);
1358 skb_queue_head_init(&csk->backlogq);
1359 skb_queue_head_init(&csk->skbq);
1360 cxgbit_sock_reset_wr_list(csk);
1361 spin_lock_init(&csk->lock);
1362 init_waitqueue_head(&csk->waitq);
1363 init_waitqueue_head(&csk->ack_waitq);
1364 csk->lock_owner = false;
1366 if (cxgbit_alloc_csk_skb(csk)) {
1372 cxgbit_get_cnp(cnp);
1373 cxgbit_get_cdev(cdev);
1375 spin_lock(&cdev->cskq.lock);
1376 list_add_tail(&csk->list, &cdev->cskq.list);
1377 spin_unlock(&cdev->cskq.lock);
1378 cxgb4_insert_tid(t, csk, tid, csk->com.local_addr.ss_family);
1379 cxgbit_pass_accept_rpl(csk, req);
1383 cxgbit_release_tid(cdev, tid);
1389 cxgbit_tx_flowc_wr_credits(struct cxgbit_sock *csk, u32 *nparamsp,
1392 u32 nparams, flowclen16, flowclen;
1394 nparams = FLOWC_WR_NPARAMS_MIN;
1396 if (csk->snd_wscale)
1399 #ifdef CONFIG_CHELSIO_T4_DCB
1402 flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
1403 flowclen16 = DIV_ROUND_UP(flowclen, 16);
1404 flowclen = flowclen16 * 16;
1406 * Return the number of 16-byte credits used by the flowc request.
1407 * Pass back the nparams and actual flowc length if requested.
1410 *nparamsp = nparams;
1412 *flowclenp = flowclen;
1416 u32 cxgbit_send_tx_flowc_wr(struct cxgbit_sock *csk)
1418 struct cxgbit_device *cdev = csk->com.cdev;
1419 struct fw_flowc_wr *flowc;
1420 u32 nparams, flowclen16, flowclen;
1421 struct sk_buff *skb;
1424 #ifdef CONFIG_CHELSIO_T4_DCB
1425 u16 vlan = ((struct l2t_entry *)csk->l2t)->vlan;
1428 flowclen16 = cxgbit_tx_flowc_wr_credits(csk, &nparams, &flowclen);
1430 skb = __skb_dequeue(&csk->skbq);
1431 flowc = __skb_put_zero(skb, flowclen);
1433 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
1434 FW_FLOWC_WR_NPARAMS_V(nparams));
1435 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
1436 FW_WR_FLOWID_V(csk->tid));
1437 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
1438 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
1439 (csk->com.cdev->lldi.pf));
1440 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
1441 flowc->mnemval[1].val = cpu_to_be32(csk->tx_chan);
1442 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
1443 flowc->mnemval[2].val = cpu_to_be32(csk->tx_chan);
1444 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
1445 flowc->mnemval[3].val = cpu_to_be32(csk->rss_qid);
1446 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
1447 flowc->mnemval[4].val = cpu_to_be32(csk->snd_nxt);
1448 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
1449 flowc->mnemval[5].val = cpu_to_be32(csk->rcv_nxt);
1450 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
1451 flowc->mnemval[6].val = cpu_to_be32(csk->snd_win);
1452 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
1453 flowc->mnemval[7].val = cpu_to_be32(csk->emss);
1455 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_TXDATAPLEN_MAX;
1456 if (test_bit(CDEV_ISO_ENABLE, &cdev->flags))
1457 flowc->mnemval[8].val = cpu_to_be32(CXGBIT_MAX_ISO_PAYLOAD);
1459 flowc->mnemval[8].val = cpu_to_be32(16384);
1463 if (csk->snd_wscale) {
1464 flowc->mnemval[index].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
1465 flowc->mnemval[index].val = cpu_to_be32(csk->snd_wscale);
1469 #ifdef CONFIG_CHELSIO_T4_DCB
1470 flowc->mnemval[index].mnemonic = FW_FLOWC_MNEM_DCBPRIO;
1471 if (vlan == VLAN_NONE) {
1472 pr_warn("csk %u without VLAN Tag on DCB Link\n", csk->tid);
1473 flowc->mnemval[index].val = cpu_to_be32(0);
1475 flowc->mnemval[index].val = cpu_to_be32(
1476 (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT);
1479 pr_debug("%s: csk %p; tx_chan = %u; rss_qid = %u; snd_seq = %u;"
1480 " rcv_seq = %u; snd_win = %u; emss = %u\n",
1481 __func__, csk, csk->tx_chan, csk->rss_qid, csk->snd_nxt,
1482 csk->rcv_nxt, csk->snd_win, csk->emss);
1483 set_wr_txq(skb, CPL_PRIORITY_DATA, csk->txq_idx);
1484 cxgbit_ofld_send(csk->com.cdev, skb);
1488 int cxgbit_setup_conn_digest(struct cxgbit_sock *csk)
1490 struct sk_buff *skb;
1491 struct cpl_set_tcb_field *req;
1492 u8 hcrc = csk->submode & CXGBIT_SUBMODE_HCRC;
1493 u8 dcrc = csk->submode & CXGBIT_SUBMODE_DCRC;
1494 unsigned int len = roundup(sizeof(*req), 16);
1497 skb = alloc_skb(len, GFP_KERNEL);
1501 /* set up ulp submode */
1502 req = __skb_put_zero(skb, len);
1504 INIT_TP_WR(req, csk->tid);
1505 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, csk->tid));
1506 req->reply_ctrl = htons(NO_REPLY_V(0) | QUEUENO_V(csk->rss_qid));
1507 req->word_cookie = htons(0);
1508 req->mask = cpu_to_be64(0x3 << 4);
1509 req->val = cpu_to_be64(((hcrc ? ULP_CRC_HEADER : 0) |
1510 (dcrc ? ULP_CRC_DATA : 0)) << 4);
1511 set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->ctrlq_idx);
1513 cxgbit_get_csk(csk);
1514 cxgbit_init_wr_wait(&csk->com.wr_wait);
1516 cxgbit_ofld_send(csk->com.cdev, skb);
1518 ret = cxgbit_wait_for_reply(csk->com.cdev,
1520 csk->tid, 5, __func__);
1527 int cxgbit_setup_conn_pgidx(struct cxgbit_sock *csk, u32 pg_idx)
1529 struct sk_buff *skb;
1530 struct cpl_set_tcb_field *req;
1531 unsigned int len = roundup(sizeof(*req), 16);
1534 skb = alloc_skb(len, GFP_KERNEL);
1538 req = __skb_put_zero(skb, len);
1540 INIT_TP_WR(req, csk->tid);
1541 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, csk->tid));
1542 req->reply_ctrl = htons(NO_REPLY_V(0) | QUEUENO_V(csk->rss_qid));
1543 req->word_cookie = htons(0);
1544 req->mask = cpu_to_be64(0x3 << 8);
1545 req->val = cpu_to_be64(pg_idx << 8);
1546 set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->ctrlq_idx);
1548 cxgbit_get_csk(csk);
1549 cxgbit_init_wr_wait(&csk->com.wr_wait);
1551 cxgbit_ofld_send(csk->com.cdev, skb);
1553 ret = cxgbit_wait_for_reply(csk->com.cdev,
1555 csk->tid, 5, __func__);
1563 cxgbit_pass_open_rpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1565 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1566 struct tid_info *t = cdev->lldi.tids;
1567 unsigned int stid = GET_TID(rpl);
1568 struct cxgbit_np *cnp = lookup_stid(t, stid);
1570 pr_debug("%s: cnp = %p; stid = %u; status = %d\n",
1571 __func__, cnp, stid, rpl->status);
1574 pr_info("%s stid %d lookup failure\n", __func__, stid);
1578 cxgbit_wake_up(&cnp->com.wr_wait, __func__, rpl->status);
1579 cxgbit_put_cnp(cnp);
1585 cxgbit_close_listsrv_rpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1587 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1588 struct tid_info *t = cdev->lldi.tids;
1589 unsigned int stid = GET_TID(rpl);
1590 struct cxgbit_np *cnp = lookup_stid(t, stid);
1592 pr_debug("%s: cnp = %p; stid = %u; status = %d\n",
1593 __func__, cnp, stid, rpl->status);
1596 pr_info("%s stid %d lookup failure\n", __func__, stid);
1600 cxgbit_wake_up(&cnp->com.wr_wait, __func__, rpl->status);
1601 cxgbit_put_cnp(cnp);
1607 cxgbit_pass_establish(struct cxgbit_device *cdev, struct sk_buff *skb)
1609 struct cpl_pass_establish *req = cplhdr(skb);
1610 struct tid_info *t = cdev->lldi.tids;
1611 unsigned int tid = GET_TID(req);
1612 struct cxgbit_sock *csk;
1613 struct cxgbit_np *cnp;
1614 u16 tcp_opt = be16_to_cpu(req->tcp_opt);
1615 u32 snd_isn = be32_to_cpu(req->snd_isn);
1616 u32 rcv_isn = be32_to_cpu(req->rcv_isn);
1618 csk = lookup_tid(t, tid);
1619 if (unlikely(!csk)) {
1620 pr_err("can't find connection for tid %u.\n", tid);
1625 pr_debug("%s: csk %p; tid %u; cnp %p\n",
1626 __func__, csk, tid, cnp);
1628 csk->write_seq = snd_isn;
1629 csk->snd_una = snd_isn;
1630 csk->snd_nxt = snd_isn;
1632 csk->rcv_nxt = rcv_isn;
1634 if (csk->rcv_win > (RCV_BUFSIZ_M << 10))
1635 csk->rx_credits = (csk->rcv_win - (RCV_BUFSIZ_M << 10));
1637 csk->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1638 cxgbit_set_emss(csk, tcp_opt);
1639 dst_confirm(csk->dst);
1640 csk->com.state = CSK_STATE_ESTABLISHED;
1641 spin_lock_bh(&cnp->np_accept_lock);
1642 list_add_tail(&csk->accept_node, &cnp->np_accept_list);
1643 spin_unlock_bh(&cnp->np_accept_lock);
1644 complete(&cnp->accept_comp);
1649 static void cxgbit_queue_rx_skb(struct cxgbit_sock *csk, struct sk_buff *skb)
1651 cxgbit_skcb_flags(skb) = 0;
1652 spin_lock_bh(&csk->rxq.lock);
1653 __skb_queue_tail(&csk->rxq, skb);
1654 spin_unlock_bh(&csk->rxq.lock);
1655 wake_up(&csk->waitq);
1658 static void cxgbit_peer_close(struct cxgbit_sock *csk, struct sk_buff *skb)
1660 pr_debug("%s: csk %p; tid %u; state %d\n",
1661 __func__, csk, csk->tid, csk->com.state);
1663 switch (csk->com.state) {
1664 case CSK_STATE_ESTABLISHED:
1665 csk->com.state = CSK_STATE_CLOSING;
1666 cxgbit_queue_rx_skb(csk, skb);
1668 case CSK_STATE_CLOSING:
1669 /* simultaneous close */
1670 csk->com.state = CSK_STATE_MORIBUND;
1672 case CSK_STATE_MORIBUND:
1673 csk->com.state = CSK_STATE_DEAD;
1674 cxgbit_put_csk(csk);
1676 case CSK_STATE_ABORTING:
1679 pr_info("%s: cpl_peer_close in bad state %d\n",
1680 __func__, csk->com.state);
1686 static void cxgbit_close_con_rpl(struct cxgbit_sock *csk, struct sk_buff *skb)
1688 pr_debug("%s: csk %p; tid %u; state %d\n",
1689 __func__, csk, csk->tid, csk->com.state);
1691 switch (csk->com.state) {
1692 case CSK_STATE_CLOSING:
1693 csk->com.state = CSK_STATE_MORIBUND;
1695 case CSK_STATE_MORIBUND:
1696 csk->com.state = CSK_STATE_DEAD;
1697 cxgbit_put_csk(csk);
1699 case CSK_STATE_ABORTING:
1700 case CSK_STATE_DEAD:
1703 pr_info("%s: cpl_close_con_rpl in bad state %d\n",
1704 __func__, csk->com.state);
1710 static void cxgbit_abort_req_rss(struct cxgbit_sock *csk, struct sk_buff *skb)
1712 struct cpl_abort_req_rss *hdr = cplhdr(skb);
1713 unsigned int tid = GET_TID(hdr);
1714 struct sk_buff *rpl_skb;
1715 bool release = false;
1716 bool wakeup_thread = false;
1717 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
1719 pr_debug("%s: csk %p; tid %u; state %d\n",
1720 __func__, csk, tid, csk->com.state);
1722 if (cxgb_is_neg_adv(hdr->status)) {
1723 pr_err("%s: got neg advise %d on tid %u\n",
1724 __func__, hdr->status, tid);
1728 switch (csk->com.state) {
1729 case CSK_STATE_CONNECTING:
1730 case CSK_STATE_MORIBUND:
1731 csk->com.state = CSK_STATE_DEAD;
1734 case CSK_STATE_ESTABLISHED:
1735 csk->com.state = CSK_STATE_DEAD;
1736 wakeup_thread = true;
1738 case CSK_STATE_CLOSING:
1739 csk->com.state = CSK_STATE_DEAD;
1743 case CSK_STATE_ABORTING:
1746 pr_info("%s: cpl_abort_req_rss in bad state %d\n",
1747 __func__, csk->com.state);
1748 csk->com.state = CSK_STATE_DEAD;
1751 __skb_queue_purge(&csk->txq);
1753 if (!test_and_set_bit(CSK_TX_DATA_SENT, &csk->com.flags))
1754 cxgbit_send_tx_flowc_wr(csk);
1756 rpl_skb = __skb_dequeue(&csk->skbq);
1758 cxgb_mk_abort_rpl(rpl_skb, len, csk->tid, csk->txq_idx);
1759 cxgbit_ofld_send(csk->com.cdev, rpl_skb);
1761 if (wakeup_thread) {
1762 cxgbit_queue_rx_skb(csk, skb);
1767 cxgbit_put_csk(csk);
1772 static void cxgbit_abort_rpl_rss(struct cxgbit_sock *csk, struct sk_buff *skb)
1774 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1776 pr_debug("%s: csk %p; tid %u; state %d\n",
1777 __func__, csk, csk->tid, csk->com.state);
1779 switch (csk->com.state) {
1780 case CSK_STATE_ABORTING:
1781 csk->com.state = CSK_STATE_DEAD;
1782 if (test_bit(CSK_ABORT_RPL_WAIT, &csk->com.flags))
1783 cxgbit_wake_up(&csk->com.wr_wait, __func__,
1785 cxgbit_put_csk(csk);
1788 pr_info("%s: cpl_abort_rpl_rss in state %d\n",
1789 __func__, csk->com.state);
1795 static bool cxgbit_credit_err(const struct cxgbit_sock *csk)
1797 const struct sk_buff *skb = csk->wr_pending_head;
1800 if (unlikely(csk->wr_cred > csk->wr_max_cred)) {
1801 pr_err("csk 0x%p, tid %u, credit %u > %u\n",
1802 csk, csk->tid, csk->wr_cred, csk->wr_max_cred);
1807 credit += (__force u32)skb->csum;
1808 skb = cxgbit_skcb_tx_wr_next(skb);
1811 if (unlikely((csk->wr_cred + credit) != csk->wr_max_cred)) {
1812 pr_err("csk 0x%p, tid %u, credit %u + %u != %u.\n",
1813 csk, csk->tid, csk->wr_cred,
1814 credit, csk->wr_max_cred);
1822 static void cxgbit_fw4_ack(struct cxgbit_sock *csk, struct sk_buff *skb)
1824 struct cpl_fw4_ack *rpl = (struct cpl_fw4_ack *)cplhdr(skb);
1825 u32 credits = rpl->credits;
1826 u32 snd_una = ntohl(rpl->snd_una);
1828 csk->wr_cred += credits;
1829 if (csk->wr_una_cred > (csk->wr_max_cred - csk->wr_cred))
1830 csk->wr_una_cred = csk->wr_max_cred - csk->wr_cred;
1833 struct sk_buff *p = cxgbit_sock_peek_wr(csk);
1837 pr_err("csk 0x%p,%u, cr %u,%u+%u, empty.\n",
1838 csk, csk->tid, credits,
1839 csk->wr_cred, csk->wr_una_cred);
1843 csum = (__force u32)p->csum;
1844 if (unlikely(credits < csum)) {
1845 pr_warn("csk 0x%p,%u, cr %u,%u+%u, < %u.\n",
1847 credits, csk->wr_cred, csk->wr_una_cred,
1849 p->csum = (__force __wsum)(csum - credits);
1853 cxgbit_sock_dequeue_wr(csk);
1858 if (unlikely(cxgbit_credit_err(csk))) {
1859 cxgbit_queue_rx_skb(csk, skb);
1863 if (rpl->seq_vld & CPL_FW4_ACK_FLAGS_SEQVAL) {
1864 if (unlikely(before(snd_una, csk->snd_una))) {
1865 pr_warn("csk 0x%p,%u, snd_una %u/%u.",
1866 csk, csk->tid, snd_una,
1871 if (csk->snd_una != snd_una) {
1872 csk->snd_una = snd_una;
1873 dst_confirm(csk->dst);
1874 wake_up(&csk->ack_waitq);
1878 if (skb_queue_len(&csk->txq))
1879 cxgbit_push_tx_frames(csk);
1885 static void cxgbit_set_tcb_rpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1887 struct cxgbit_sock *csk;
1888 struct cpl_set_tcb_rpl *rpl = (struct cpl_set_tcb_rpl *)skb->data;
1889 unsigned int tid = GET_TID(rpl);
1890 struct cxgb4_lld_info *lldi = &cdev->lldi;
1891 struct tid_info *t = lldi->tids;
1893 csk = lookup_tid(t, tid);
1894 if (unlikely(!csk)) {
1895 pr_err("can't find connection for tid %u.\n", tid);
1898 cxgbit_wake_up(&csk->com.wr_wait, __func__, rpl->status);
1901 cxgbit_put_csk(csk);
1906 static void cxgbit_rx_data(struct cxgbit_device *cdev, struct sk_buff *skb)
1908 struct cxgbit_sock *csk;
1909 struct cpl_rx_data *cpl = cplhdr(skb);
1910 unsigned int tid = GET_TID(cpl);
1911 struct cxgb4_lld_info *lldi = &cdev->lldi;
1912 struct tid_info *t = lldi->tids;
1914 csk = lookup_tid(t, tid);
1915 if (unlikely(!csk)) {
1916 pr_err("can't find conn. for tid %u.\n", tid);
1920 cxgbit_queue_rx_skb(csk, skb);
1927 __cxgbit_process_rx_cpl(struct cxgbit_sock *csk, struct sk_buff *skb)
1929 spin_lock(&csk->lock);
1930 if (csk->lock_owner) {
1931 __skb_queue_tail(&csk->backlogq, skb);
1932 spin_unlock(&csk->lock);
1936 cxgbit_skcb_rx_backlog_fn(skb)(csk, skb);
1937 spin_unlock(&csk->lock);
1940 static void cxgbit_process_rx_cpl(struct cxgbit_sock *csk, struct sk_buff *skb)
1942 cxgbit_get_csk(csk);
1943 __cxgbit_process_rx_cpl(csk, skb);
1944 cxgbit_put_csk(csk);
1947 static void cxgbit_rx_cpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1949 struct cxgbit_sock *csk;
1950 struct cpl_tx_data *cpl = cplhdr(skb);
1951 struct cxgb4_lld_info *lldi = &cdev->lldi;
1952 struct tid_info *t = lldi->tids;
1953 unsigned int tid = GET_TID(cpl);
1954 u8 opcode = cxgbit_skcb_rx_opcode(skb);
1959 cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_fw4_ack;
1962 case CPL_PEER_CLOSE:
1963 cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_peer_close;
1965 case CPL_CLOSE_CON_RPL:
1966 cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_close_con_rpl;
1968 case CPL_ABORT_REQ_RSS:
1969 cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_abort_req_rss;
1971 case CPL_ABORT_RPL_RSS:
1972 cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_abort_rpl_rss;
1978 csk = lookup_tid(t, tid);
1979 if (unlikely(!csk)) {
1980 pr_err("can't find conn. for tid %u.\n", tid);
1985 cxgbit_process_rx_cpl(csk, skb);
1987 __cxgbit_process_rx_cpl(csk, skb);
1994 cxgbit_cplhandler_func cxgbit_cplhandlers[NUM_CPL_CMDS] = {
1995 [CPL_PASS_OPEN_RPL] = cxgbit_pass_open_rpl,
1996 [CPL_CLOSE_LISTSRV_RPL] = cxgbit_close_listsrv_rpl,
1997 [CPL_PASS_ACCEPT_REQ] = cxgbit_pass_accept_req,
1998 [CPL_PASS_ESTABLISH] = cxgbit_pass_establish,
1999 [CPL_SET_TCB_RPL] = cxgbit_set_tcb_rpl,
2000 [CPL_RX_DATA] = cxgbit_rx_data,
2001 [CPL_FW4_ACK] = cxgbit_rx_cpl,
2002 [CPL_PEER_CLOSE] = cxgbit_rx_cpl,
2003 [CPL_CLOSE_CON_RPL] = cxgbit_rx_cpl,
2004 [CPL_ABORT_REQ_RSS] = cxgbit_rx_cpl,
2005 [CPL_ABORT_RPL_RSS] = cxgbit_rx_cpl,