Index: head/sys/dev/cxgbe/iw_cxgbe/cm.c =================================================================== --- head/sys/dev/cxgbe/iw_cxgbe/cm.c (revision 340062) +++ head/sys/dev/cxgbe/iw_cxgbe/cm.c (revision 340063) @@ -1,2986 +1,3011 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009-2013, 2016 Chelsio, Inc. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU * General Public License (GPL) Version 2, available from the file * COPYING in the main directory of this source tree, or the * OpenIB.org BSD license below: * * Redistribution and use in source and binary forms, with or * without modification, are permitted provided that the following * conditions are met: * * - Redistributions of source code must retain the above * copyright notice, this list of conditions and the following * disclaimer. * * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials * provided with the distribution. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #ifdef TCP_OFFLOAD #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include struct sge_iq; struct rss_header; struct cpl_set_tcb_rpl; #include #include "offload.h" #include "tom/t4_tom.h" #define TOEPCB(so) ((struct toepcb *)(so_sototcpcb((so))->t_toe)) #include "iw_cxgbe.h" #include #include #include #include #include #include #include static spinlock_t req_lock; static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list; static struct work_struct c4iw_task; static struct workqueue_struct *c4iw_taskq; static LIST_HEAD(err_cqe_list); static spinlock_t err_cqe_lock; static LIST_HEAD(listen_port_list); static DEFINE_MUTEX(listen_port_mutex); static void process_req(struct work_struct *ctx); static void start_ep_timer(struct c4iw_ep *ep); static int stop_ep_timer(struct c4iw_ep *ep); static int set_tcpinfo(struct c4iw_ep *ep); static void process_timeout(struct c4iw_ep *ep); static void process_err_cqes(void); static void *alloc_ep(int size, gfp_t flags); static void close_socket(struct socket *so); static int send_mpa_req(struct c4iw_ep *ep); static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen); static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen); static void close_complete_upcall(struct c4iw_ep *ep, int status); static int send_abort(struct c4iw_ep *ep); static void peer_close_upcall(struct c4iw_ep *ep); static void peer_abort_upcall(struct c4iw_ep *ep); static void connect_reply_upcall(struct c4iw_ep *ep, int status); static int connect_request_upcall(struct c4iw_ep *ep); static void established_upcall(struct c4iw_ep *ep); static int process_mpa_reply(struct c4iw_ep *ep); static int process_mpa_request(struct c4iw_ep *ep); static void process_peer_close(struct c4iw_ep *ep); static void process_conn_error(struct c4iw_ep *ep); static void process_close_complete(struct c4iw_ep *ep); static void ep_timeout(unsigned long arg); static void setiwsockopt(struct socket *so); static void init_iwarp_socket(struct socket *so, void *arg); static void uninit_iwarp_socket(struct socket *so); static void process_data(struct c4iw_ep *ep); static void process_connected(struct c4iw_ep *ep); static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag); static void process_socket_event(struct c4iw_ep *ep); static void release_ep_resources(struct c4iw_ep *ep); static int process_terminate(struct c4iw_ep *ep); static int terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m); static int add_ep_to_req_list(struct c4iw_ep *ep, int ep_events); static struct listen_port_info * add_ep_to_listenlist(struct c4iw_listen_ep *lep); static int rem_ep_from_listenlist(struct c4iw_listen_ep *lep); static struct c4iw_listen_ep * find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so); static int get_ifnet_from_raddr(struct sockaddr_storage *raddr, struct ifnet **ifp); static void process_newconn(struct c4iw_listen_ep *master_lep, struct socket *new_so); #define START_EP_TIMER(ep) \ do { \ CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \ __func__, __LINE__, (ep)); \ start_ep_timer(ep); \ } while (0) #define STOP_EP_TIMER(ep) \ ({ \ CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \ __func__, __LINE__, (ep)); \ stop_ep_timer(ep); \ }) #define GET_LOCAL_ADDR(pladdr, so) \ do { \ struct sockaddr_storage *__a = NULL; \ struct inpcb *__inp = sotoinpcb(so); \ KASSERT(__inp != NULL, \ ("GET_LOCAL_ADDR(%s):so:%p, inp = NULL", __func__, so)); \ if (__inp->inp_vflag & INP_IPV4) \ in_getsockaddr(so, (struct sockaddr **)&__a); \ else \ in6_getsockaddr(so, (struct sockaddr **)&__a); \ *(pladdr) = *__a; \ free(__a, M_SONAME); \ } while (0) #define GET_REMOTE_ADDR(praddr, so) \ do { \ struct sockaddr_storage *__a = NULL; \ struct inpcb *__inp = sotoinpcb(so); \ KASSERT(__inp != NULL, \ ("GET_REMOTE_ADDR(%s):so:%p, inp = NULL", __func__, so)); \ if (__inp->inp_vflag & INP_IPV4) \ in_getpeeraddr(so, (struct sockaddr **)&__a); \ else \ in6_getpeeraddr(so, (struct sockaddr **)&__a); \ *(praddr) = *__a; \ free(__a, M_SONAME); \ } while (0) -#ifdef KTR static char *states[] = { "idle", "listen", "connecting", "mpa_wait_req", "mpa_req_sent", "mpa_req_rcvd", "mpa_rep_sent", "fpdu_mode", "aborting", "closing", "moribund", "dead", NULL, }; -#endif static void deref_cm_id(struct c4iw_ep_common *epc) { epc->cm_id->rem_ref(epc->cm_id); epc->cm_id = NULL; set_bit(CM_ID_DEREFED, &epc->history); } static void ref_cm_id(struct c4iw_ep_common *epc) { set_bit(CM_ID_REFED, &epc->history); epc->cm_id->add_ref(epc->cm_id); } static void deref_qp(struct c4iw_ep *ep) { c4iw_qp_rem_ref(&ep->com.qp->ibqp); clear_bit(QP_REFERENCED, &ep->com.flags); set_bit(QP_DEREFED, &ep->com.history); } static void ref_qp(struct c4iw_ep *ep) { set_bit(QP_REFERENCED, &ep->com.flags); set_bit(QP_REFED, &ep->com.history); c4iw_qp_add_ref(&ep->com.qp->ibqp); } /* allocated per TCP port while listening */ struct listen_port_info { uint16_t port_num; /* TCP port address */ struct list_head list; /* belongs to listen_port_list */ struct list_head lep_list; /* per port lep list */ uint32_t refcnt; /* number of lep's listening */ }; /* * Following two lists are used to manage INADDR_ANY listeners: * 1)listen_port_list * 2)lep_list * * Below is the INADDR_ANY listener lists overview on a system with a two port * adapter: * |------------------| * |listen_port_list | * |------------------| * | * | |-----------| |-----------| * | | port_num:X| | port_num:X| * |--------------|-list------|-------|-list------|-------.... * | lep_list----| | lep_list----| * | refcnt | | | refcnt | | * | | | | | | * | | | | | | * |-----------| | |-----------| | * | | * | | * | | * | | lep1 lep2 * | | |----------------| |----------------| * | |----| listen_ep_list |----| listen_ep_list | * | |----------------| |----------------| * | * | * | lep1 lep2 * | |----------------| |----------------| * |---| listen_ep_list |----| listen_ep_list | * |----------------| |----------------| * * Because of two port adapter, the number of lep's are two(lep1 & lep2) for * each TCP port number. * * Here 'lep1' is always marked as Master lep, because solisten() is always * called through first lep. * */ static struct listen_port_info * add_ep_to_listenlist(struct c4iw_listen_ep *lep) { uint16_t port; struct listen_port_info *port_info = NULL; struct sockaddr_storage *laddr = &lep->com.local_addr; port = (laddr->ss_family == AF_INET) ? ((struct sockaddr_in *)laddr)->sin_port : ((struct sockaddr_in6 *)laddr)->sin6_port; mutex_lock(&listen_port_mutex); list_for_each_entry(port_info, &listen_port_list, list) if (port_info->port_num == port) goto found_port; port_info = malloc(sizeof(*port_info), M_CXGBE, M_WAITOK); port_info->port_num = port; port_info->refcnt = 0; list_add_tail(&port_info->list, &listen_port_list); INIT_LIST_HEAD(&port_info->lep_list); found_port: port_info->refcnt++; list_add_tail(&lep->listen_ep_list, &port_info->lep_list); mutex_unlock(&listen_port_mutex); return port_info; } static int rem_ep_from_listenlist(struct c4iw_listen_ep *lep) { uint16_t port; struct listen_port_info *port_info = NULL; struct sockaddr_storage *laddr = &lep->com.local_addr; int refcnt = 0; port = (laddr->ss_family == AF_INET) ? ((struct sockaddr_in *)laddr)->sin_port : ((struct sockaddr_in6 *)laddr)->sin6_port; mutex_lock(&listen_port_mutex); /* get the port_info structure based on the lep's port address */ list_for_each_entry(port_info, &listen_port_list, list) { if (port_info->port_num == port) { port_info->refcnt--; refcnt = port_info->refcnt; /* remove the current lep from the listen list */ list_del(&lep->listen_ep_list); if (port_info->refcnt == 0) { /* Remove this entry from the list as there * are no more listeners for this port_num. */ list_del(&port_info->list); kfree(port_info); } break; } } mutex_unlock(&listen_port_mutex); return refcnt; } /* * Find the lep that belongs to the ifnet on which the SYN frame was received. */ struct c4iw_listen_ep * find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so) { struct adapter *adap = NULL; struct c4iw_listen_ep *lep = NULL; struct ifnet *ifp = NULL, *hw_ifp = NULL; struct listen_port_info *port_info = NULL; int i = 0, found_portinfo = 0, found_lep = 0; uint16_t port; /* * STEP 1: Figure out 'ifp' of the physical interface, not pseudo * interfaces like vlan, lagg, etc.. * TBD: lagg support, lagg + vlan support. */ ifp = TOEPCB(so)->l2te->ifp; if (ifp->if_type == IFT_L2VLAN) { hw_ifp = VLAN_TRUNKDEV(ifp); if (hw_ifp == NULL) { CTR4(KTR_IW_CXGBE, "%s: Failed to get parent ifnet of " "vlan ifnet %p, sock %p, master_lep %p", __func__, ifp, so, master_lep); return (NULL); } } else hw_ifp = ifp; /* STEP 2: Find 'port_info' with listener local port address. */ port = (master_lep->com.local_addr.ss_family == AF_INET) ? ((struct sockaddr_in *)&master_lep->com.local_addr)->sin_port : ((struct sockaddr_in6 *)&master_lep->com.local_addr)->sin6_port; mutex_lock(&listen_port_mutex); list_for_each_entry(port_info, &listen_port_list, list) if (port_info->port_num == port) { found_portinfo =1; break; } if (!found_portinfo) goto out; /* STEP 3: Traverse through list of lep's that are bound to the current * TCP port address and find the lep that belongs to the ifnet on which * the SYN frame was received. */ list_for_each_entry(lep, &port_info->lep_list, listen_ep_list) { adap = lep->com.dev->rdev.adap; for_each_port(adap, i) { if (hw_ifp == adap->port[i]->vi[0].ifp) { found_lep =1; goto out; } } } out: mutex_unlock(&listen_port_mutex); return found_lep ? lep : (NULL); } static void process_timeout(struct c4iw_ep *ep) { struct c4iw_qp_attributes attrs = {0}; int abort = 1; CTR4(KTR_IW_CXGBE, "%s ep :%p, tid:%u, state %d", __func__, ep, ep->hwtid, ep->com.state); set_bit(TIMEDOUT, &ep->com.history); switch (ep->com.state) { case MPA_REQ_SENT: connect_reply_upcall(ep, -ETIMEDOUT); break; case MPA_REQ_WAIT: case MPA_REQ_RCVD: case MPA_REP_SENT: case FPDU_MODE: break; case CLOSING: case MORIBUND: if (ep->com.cm_id && ep->com.qp) { attrs.next_state = C4IW_QP_STATE_ERROR; c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); } close_complete_upcall(ep, -ETIMEDOUT); break; case ABORTING: case DEAD: /* * These states are expected if the ep timed out at the same * time as another thread was calling stop_ep_timer(). * So we silently do nothing for these states. */ abort = 0; break; default: CTR4(KTR_IW_CXGBE, "%s unexpected state ep %p tid %u state %u\n" , __func__, ep, ep->hwtid, ep->com.state); abort = 0; } if (abort) c4iw_ep_disconnect(ep, 1, GFP_KERNEL); c4iw_put_ep(&ep->com); return; } struct cqe_list_entry { struct list_head entry; struct c4iw_dev *rhp; struct t4_cqe err_cqe; }; static void process_err_cqes(void) { unsigned long flag; struct cqe_list_entry *cle; spin_lock_irqsave(&err_cqe_lock, flag); while (!list_empty(&err_cqe_list)) { struct list_head *tmp; tmp = err_cqe_list.next; list_del(tmp); tmp->next = tmp->prev = NULL; spin_unlock_irqrestore(&err_cqe_lock, flag); cle = list_entry(tmp, struct cqe_list_entry, entry); c4iw_ev_dispatch(cle->rhp, &cle->err_cqe); free(cle, M_CXGBE); spin_lock_irqsave(&err_cqe_lock, flag); } spin_unlock_irqrestore(&err_cqe_lock, flag); return; } static void process_req(struct work_struct *ctx) { struct c4iw_ep_common *epc; unsigned long flag; int ep_events; process_err_cqes(); spin_lock_irqsave(&req_lock, flag); while (!TAILQ_EMPTY(&req_list)) { epc = TAILQ_FIRST(&req_list); TAILQ_REMOVE(&req_list, epc, entry); epc->entry.tqe_prev = NULL; ep_events = epc->ep_events; epc->ep_events = 0; spin_unlock_irqrestore(&req_lock, flag); mutex_lock(&epc->mutex); CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, ep_state %s events 0x%x", __func__, epc->so, epc, states[epc->state], ep_events); if (ep_events & C4IW_EVENT_TERM) process_terminate((struct c4iw_ep *)epc); if (ep_events & C4IW_EVENT_TIMEOUT) process_timeout((struct c4iw_ep *)epc); if (ep_events & C4IW_EVENT_SOCKET) process_socket_event((struct c4iw_ep *)epc); mutex_unlock(&epc->mutex); c4iw_put_ep(epc); process_err_cqes(); spin_lock_irqsave(&req_lock, flag); } spin_unlock_irqrestore(&req_lock, flag); } /* * XXX: doesn't belong here in the iWARP driver. * XXX: assumes that the connection was offloaded by cxgbe/t4_tom if TF_TOE is * set. Is this a valid assumption for active open? */ static int set_tcpinfo(struct c4iw_ep *ep) { struct socket *so = ep->com.so; struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp; struct toepcb *toep; int rc = 0; INP_WLOCK(inp); tp = intotcpcb(inp); if ((tp->t_flags & TF_TOE) == 0) { rc = EINVAL; log(LOG_ERR, "%s: connection not offloaded (so %p, ep %p)\n", __func__, so, ep); goto done; } toep = TOEPCB(so); ep->hwtid = toep->tid; ep->snd_seq = tp->snd_nxt; ep->rcv_seq = tp->rcv_nxt; ep->emss = max(tp->t_maxseg, 128); done: INP_WUNLOCK(inp); return (rc); } static int get_ifnet_from_raddr(struct sockaddr_storage *raddr, struct ifnet **ifp) { int err = 0; if (raddr->ss_family == AF_INET) { struct sockaddr_in *raddr4 = (struct sockaddr_in *)raddr; struct nhop4_extended nh4 = {0}; err = fib4_lookup_nh_ext(RT_DEFAULT_FIB, raddr4->sin_addr, NHR_REF, 0, &nh4); *ifp = nh4.nh_ifp; if (err) fib4_free_nh_ext(RT_DEFAULT_FIB, &nh4); } else { struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)raddr; struct nhop6_extended nh6 = {0}; struct in6_addr addr6; uint32_t scopeid; memset(&addr6, 0, sizeof(addr6)); in6_splitscope((struct in6_addr *)&raddr6->sin6_addr, &addr6, &scopeid); err = fib6_lookup_nh_ext(RT_DEFAULT_FIB, &addr6, scopeid, NHR_REF, 0, &nh6); *ifp = nh6.nh_ifp; if (err) fib6_free_nh_ext(RT_DEFAULT_FIB, &nh6); } CTR2(KTR_IW_CXGBE, "%s: return: %d", __func__, err); return err; } static void close_socket(struct socket *so) { uninit_iwarp_socket(so); soclose(so); } static void process_peer_close(struct c4iw_ep *ep) { struct c4iw_qp_attributes attrs = {0}; int disconnect = 1; int release = 0; CTR4(KTR_IW_CXGBE, "%s:ppcB ep %p so %p state %s", __func__, ep, ep->com.so, states[ep->com.state]); switch (ep->com.state) { case MPA_REQ_WAIT: CTR2(KTR_IW_CXGBE, "%s:ppc1 %p MPA_REQ_WAIT DEAD", __func__, ep); /* Fallthrough */ case MPA_REQ_SENT: CTR2(KTR_IW_CXGBE, "%s:ppc2 %p MPA_REQ_SENT DEAD", __func__, ep); ep->com.state = DEAD; connect_reply_upcall(ep, -ECONNABORTED); disconnect = 0; STOP_EP_TIMER(ep); close_socket(ep->com.so); deref_cm_id(&ep->com); release = 1; break; case MPA_REQ_RCVD: /* * We're gonna mark this puppy DEAD, but keep * the reference on it until the ULP accepts or * rejects the CR. */ CTR2(KTR_IW_CXGBE, "%s:ppc3 %p MPA_REQ_RCVD CLOSING", __func__, ep); ep->com.state = CLOSING; break; case MPA_REP_SENT: CTR2(KTR_IW_CXGBE, "%s:ppc4 %p MPA_REP_SENT CLOSING", __func__, ep); ep->com.state = CLOSING; break; case FPDU_MODE: CTR2(KTR_IW_CXGBE, "%s:ppc5 %p FPDU_MODE CLOSING", __func__, ep); START_EP_TIMER(ep); ep->com.state = CLOSING; attrs.next_state = C4IW_QP_STATE_CLOSING; c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); peer_close_upcall(ep); break; case ABORTING: CTR2(KTR_IW_CXGBE, "%s:ppc6 %p ABORTING (disconn)", __func__, ep); disconnect = 0; break; case CLOSING: CTR2(KTR_IW_CXGBE, "%s:ppc7 %p CLOSING MORIBUND", __func__, ep); ep->com.state = MORIBUND; disconnect = 0; break; case MORIBUND: CTR2(KTR_IW_CXGBE, "%s:ppc8 %p MORIBUND DEAD", __func__, ep); STOP_EP_TIMER(ep); if (ep->com.cm_id && ep->com.qp) { attrs.next_state = C4IW_QP_STATE_IDLE; c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); } close_socket(ep->com.so); close_complete_upcall(ep, 0); ep->com.state = DEAD; release = 1; disconnect = 0; break; case DEAD: CTR2(KTR_IW_CXGBE, "%s:ppc9 %p DEAD (disconn)", __func__, ep); disconnect = 0; break; default: panic("%s: ep %p state %d", __func__, ep, ep->com.state); break; } if (disconnect) { CTR2(KTR_IW_CXGBE, "%s:ppca %p", __func__, ep); c4iw_ep_disconnect(ep, 0, M_NOWAIT); } if (release) { CTR2(KTR_IW_CXGBE, "%s:ppcb %p", __func__, ep); c4iw_put_ep(&ep->com); } CTR2(KTR_IW_CXGBE, "%s:ppcE %p", __func__, ep); return; } static void process_conn_error(struct c4iw_ep *ep) { struct c4iw_qp_attributes attrs = {0}; int ret; int state; state = ep->com.state; CTR5(KTR_IW_CXGBE, "%s:pceB ep %p so %p so->so_error %u state %s", __func__, ep, ep->com.so, ep->com.so->so_error, states[ep->com.state]); switch (state) { case MPA_REQ_WAIT: STOP_EP_TIMER(ep); c4iw_put_ep(&ep->parent_ep->com); break; case MPA_REQ_SENT: STOP_EP_TIMER(ep); connect_reply_upcall(ep, -ECONNRESET); break; case MPA_REP_SENT: ep->com.rpl_err = ECONNRESET; CTR1(KTR_IW_CXGBE, "waking up ep %p", ep); break; case MPA_REQ_RCVD: break; case MORIBUND: case CLOSING: STOP_EP_TIMER(ep); /*FALLTHROUGH*/ case FPDU_MODE: if (ep->com.cm_id && ep->com.qp) { attrs.next_state = C4IW_QP_STATE_ERROR; ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); if (ret) log(LOG_ERR, "%s - qp <- error failed!\n", __func__); } peer_abort_upcall(ep); break; case ABORTING: break; case DEAD: CTR2(KTR_IW_CXGBE, "%s so_error %d IN DEAD STATE!!!!", __func__, ep->com.so->so_error); return; default: panic("%s: ep %p state %d", __func__, ep, state); break; } if (state != ABORTING) { close_socket(ep->com.so); ep->com.state = DEAD; c4iw_put_ep(&ep->com); } CTR2(KTR_IW_CXGBE, "%s:pceE %p", __func__, ep); return; } static void process_close_complete(struct c4iw_ep *ep) { struct c4iw_qp_attributes attrs = {0}; int release = 0; CTR4(KTR_IW_CXGBE, "%s:pccB ep %p so %p state %s", __func__, ep, ep->com.so, states[ep->com.state]); /* The cm_id may be null if we failed to connect */ set_bit(CLOSE_CON_RPL, &ep->com.history); switch (ep->com.state) { case CLOSING: CTR2(KTR_IW_CXGBE, "%s:pcc1 %p CLOSING MORIBUND", __func__, ep); ep->com.state = MORIBUND; break; case MORIBUND: CTR2(KTR_IW_CXGBE, "%s:pcc1 %p MORIBUND DEAD", __func__, ep); STOP_EP_TIMER(ep); if ((ep->com.cm_id) && (ep->com.qp)) { CTR2(KTR_IW_CXGBE, "%s:pcc2 %p QP_STATE_IDLE", __func__, ep); attrs.next_state = C4IW_QP_STATE_IDLE; c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); } close_socket(ep->com.so); close_complete_upcall(ep, 0); ep->com.state = DEAD; release = 1; break; case ABORTING: CTR2(KTR_IW_CXGBE, "%s:pcc5 %p ABORTING", __func__, ep); break; case DEAD: CTR2(KTR_IW_CXGBE, "%s:pcc6 %p DEAD", __func__, ep); break; default: CTR2(KTR_IW_CXGBE, "%s:pcc7 %p unknown ep state", __func__, ep); panic("%s:pcc6 %p unknown ep state", __func__, ep); break; } if (release) { CTR2(KTR_IW_CXGBE, "%s:pcc8 %p", __func__, ep); release_ep_resources(ep); } CTR2(KTR_IW_CXGBE, "%s:pccE %p", __func__, ep); return; } static void setiwsockopt(struct socket *so) { int rc; struct sockopt sopt; int on = 1; sopt.sopt_dir = SOPT_SET; sopt.sopt_level = IPPROTO_TCP; sopt.sopt_name = TCP_NODELAY; sopt.sopt_val = (caddr_t)&on; sopt.sopt_valsize = sizeof on; sopt.sopt_td = NULL; rc = sosetopt(so, &sopt); if (rc) { log(LOG_ERR, "%s: can't set TCP_NODELAY on so %p (%d)\n", __func__, so, rc); } } static void init_iwarp_socket(struct socket *so, void *arg) { if (SOLISTENING(so)) { SOLISTEN_LOCK(so); solisten_upcall_set(so, c4iw_so_upcall, arg); so->so_state |= SS_NBIO; SOLISTEN_UNLOCK(so); } else { SOCKBUF_LOCK(&so->so_rcv); soupcall_set(so, SO_RCV, c4iw_so_upcall, arg); so->so_state |= SS_NBIO; SOCKBUF_UNLOCK(&so->so_rcv); } } static void uninit_iwarp_socket(struct socket *so) { if (SOLISTENING(so)) { SOLISTEN_LOCK(so); solisten_upcall_set(so, NULL, NULL); SOLISTEN_UNLOCK(so); } else { SOCKBUF_LOCK(&so->so_rcv); soupcall_clear(so, SO_RCV); SOCKBUF_UNLOCK(&so->so_rcv); } } static void process_data(struct c4iw_ep *ep) { + int ret = 0; int disconnect = 0; + struct c4iw_qp_attributes attrs = {0}; CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sbused %d", __func__, ep->com.so, ep, states[ep->com.state], sbused(&ep->com.so->so_rcv)); switch (ep->com.state) { case MPA_REQ_SENT: disconnect = process_mpa_reply(ep); break; case MPA_REQ_WAIT: disconnect = process_mpa_request(ep); if (disconnect) /* Refered in process_newconn() */ c4iw_put_ep(&ep->parent_ep->com); break; + case FPDU_MODE: + MPASS(ep->com.qp != NULL); + attrs.next_state = C4IW_QP_STATE_TERMINATE; + ret = c4iw_modify_qp(ep->com.dev, ep->com.qp, + C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); + if (ret != -EINPROGRESS) + disconnect = 1; + break; default: - if (sbused(&ep->com.so->so_rcv)) - log(LOG_ERR, "%s: Unexpected streaming data. ep %p, " + log(LOG_ERR, "%s: Unexpected streaming data. ep %p, " "state %d, so %p, so_state 0x%x, sbused %u\n", __func__, ep, ep->com.state, ep->com.so, ep->com.so->so_state, sbused(&ep->com.so->so_rcv)); break; } if (disconnect) c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL); } static void process_connected(struct c4iw_ep *ep) { struct socket *so = ep->com.so; if ((so->so_state & SS_ISCONNECTED) && !so->so_error) { if (send_mpa_req(ep)) goto err; } else { connect_reply_upcall(ep, -so->so_error); goto err; } return; err: close_socket(so); ep->com.state = DEAD; c4iw_put_ep(&ep->com); return; } static inline int c4iw_zero_addr(struct sockaddr *addr) { struct in6_addr *ip6; if (addr->sa_family == AF_INET) return IN_ZERONET( ntohl(((struct sockaddr_in *)addr)->sin_addr.s_addr)); else { ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr; return (ip6->s6_addr32[0] | ip6->s6_addr32[1] | ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0; } } static inline int c4iw_loopback_addr(struct sockaddr *addr) { if (addr->sa_family == AF_INET) return IN_LOOPBACK( ntohl(((struct sockaddr_in *) addr)->sin_addr.s_addr)); else return IN6_IS_ADDR_LOOPBACK( &((struct sockaddr_in6 *) addr)->sin6_addr); } static inline int c4iw_any_addr(struct sockaddr *addr) { return c4iw_zero_addr(addr) || c4iw_loopback_addr(addr); } static void process_newconn(struct c4iw_listen_ep *master_lep, struct socket *new_so) { struct c4iw_listen_ep *real_lep = NULL; struct c4iw_ep *new_ep = NULL; struct sockaddr_in *remote = NULL; int ret = 0; MPASS(new_so != NULL); if (c4iw_any_addr((struct sockaddr *)&master_lep->com.local_addr)) { /* Here we need to find the 'real_lep' that belongs to the * incomming socket's network interface, such that the newly * created 'ep' can be attached to the real 'lep'. */ real_lep = find_real_listen_ep(master_lep, new_so); if (real_lep == NULL) { CTR2(KTR_IW_CXGBE, "%s: Could not find the real listen " "ep for sock: %p", __func__, new_so); log(LOG_ERR,"%s: Could not find the real listen ep for " "sock: %p\n", __func__, new_so); /* FIXME: properly free the 'new_so' in failure case. * Use of soabort() and soclose() are not legal * here(before soaccept()). */ return; } } else /* for Non-Wildcard address, master_lep is always the real_lep */ real_lep = master_lep; new_ep = alloc_ep(sizeof(*new_ep), GFP_KERNEL); CTR6(KTR_IW_CXGBE, "%s: master_lep %p, real_lep: %p, new ep %p, " "listening so %p, new so %p", __func__, master_lep, real_lep, new_ep, master_lep->com.so, new_so); new_ep->com.dev = real_lep->com.dev; new_ep->com.so = new_so; new_ep->com.cm_id = NULL; new_ep->com.thread = real_lep->com.thread; new_ep->parent_ep = real_lep; GET_LOCAL_ADDR(&new_ep->com.local_addr, new_so); GET_REMOTE_ADDR(&new_ep->com.remote_addr, new_so); c4iw_get_ep(&real_lep->com); init_timer(&new_ep->timer); new_ep->com.state = MPA_REQ_WAIT; START_EP_TIMER(new_ep); setiwsockopt(new_so); ret = soaccept(new_so, (struct sockaddr **)&remote); if (ret != 0) { CTR4(KTR_IW_CXGBE, "%s:listen sock:%p, new sock:%p, ret:%d\n", __func__, master_lep->com.so, new_so, ret); if (remote != NULL) free(remote, M_SONAME); uninit_iwarp_socket(new_so); soclose(new_so); c4iw_put_ep(&new_ep->com); c4iw_put_ep(&real_lep->com); return; } free(remote, M_SONAME); /* MPA request might have been queued up on the socket already, so we * initialize the socket/upcall_handler under lock to prevent processing * MPA request on another thread(via process_req()) simultaniously. */ c4iw_get_ep(&new_ep->com); /* Dereferenced at the end below, this is to avoid freeing of ep before ep unlock. */ mutex_lock(&new_ep->com.mutex); init_iwarp_socket(new_so, &new_ep->com); ret = process_mpa_request(new_ep); if (ret) { /* ABORT */ c4iw_ep_disconnect(new_ep, 1, GFP_KERNEL); c4iw_put_ep(&real_lep->com); } mutex_unlock(&new_ep->com.mutex); c4iw_put_ep(&new_ep->com); return; } static int add_ep_to_req_list(struct c4iw_ep *ep, int new_ep_event) { unsigned long flag; spin_lock_irqsave(&req_lock, flag); if (ep && ep->com.so) { ep->com.ep_events |= new_ep_event; if (!ep->com.entry.tqe_prev) { c4iw_get_ep(&ep->com); TAILQ_INSERT_TAIL(&req_list, &ep->com, entry); queue_work(c4iw_taskq, &c4iw_task); } } spin_unlock_irqrestore(&req_lock, flag); return (0); } static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag) { struct c4iw_ep *ep = arg; CTR6(KTR_IW_CXGBE, "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p", __func__, so, so->so_state, ep, states[ep->com.state], ep->com.entry.tqe_prev); MPASS(ep->com.so == so); /* * Wake up any threads waiting in rdma_init()/rdma_fini(), * with locks held. */ if (so->so_error) c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); add_ep_to_req_list(ep, C4IW_EVENT_SOCKET); return (SU_OK); } static int terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_rdma_terminate *cpl = mtod(m, const void *); unsigned int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct socket *so; struct c4iw_ep *ep; INP_WLOCK(toep->inp); so = inp_inpcbtosocket(toep->inp); ep = so->so_rcv.sb_upcallarg; INP_WUNLOCK(toep->inp); CTR3(KTR_IW_CXGBE, "%s: so %p, ep %p", __func__, so, ep); add_ep_to_req_list(ep, C4IW_EVENT_TERM); return 0; } static void process_socket_event(struct c4iw_ep *ep) { int state = ep->com.state; struct socket *so = ep->com.so; if (ep->com.state == DEAD) { CTR3(KTR_IW_CXGBE, "%s: Pending socket event discarded " "ep %p ep_state %s", __func__, ep, states[state]); return; } CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, " "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state, so->so_error, so->so_rcv.sb_state, ep, states[state]); if (state == CONNECTING) { process_connected(ep); return; } if (state == LISTEN) { struct c4iw_listen_ep *lep = (struct c4iw_listen_ep *)ep; struct socket *listen_so = so, *new_so = NULL; int error = 0; SOLISTEN_LOCK(listen_so); do { error = solisten_dequeue(listen_so, &new_so, SOCK_NONBLOCK); if (error) { CTR4(KTR_IW_CXGBE, "%s: lep %p listen_so %p " "error %d", __func__, lep, listen_so, error); return; } process_newconn(lep, new_so); /* solisten_dequeue() unlocks while return, so aquire * lock again for sol_qlen and also for next iteration. */ SOLISTEN_LOCK(listen_so); } while (listen_so->sol_qlen); SOLISTEN_UNLOCK(listen_so); return; } /* connection error */ if (so->so_error) { process_conn_error(ep); return; } /* peer close */ if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state <= CLOSING) { process_peer_close(ep); /* * check whether socket disconnect event is pending before * returning. Fallthrough if yes. */ if (!(so->so_state & SS_ISDISCONNECTED)) return; } /* close complete */ if (so->so_state & SS_ISDISCONNECTED) { process_close_complete(ep); return; } /* rx data */ - process_data(ep); + if (sbused(&ep->com.so->so_rcv)) { + process_data(ep); + return; + } + + /* Socket events for 'MPA Request Received' and 'Close Complete' + * were already processed earlier in their previous events handlers. + * Hence, these socket events are skipped. + * And any other socket events must have handled above. + */ + MPASS((ep->com.state == MPA_REQ_RCVD) || (ep->com.state == MORIBUND)); + + if ((ep->com.state != MPA_REQ_RCVD) && (ep->com.state != MORIBUND)) + log(LOG_ERR, "%s: Unprocessed socket event so %p, " + "so_state 0x%x, so_err %d, sb_state 0x%x, ep %p, ep_state %s\n", + __func__, so, so->so_state, so->so_error, so->so_rcv.sb_state, + ep, states[state]); + } SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters"); static int dack_mode = 0; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RWTUN, &dack_mode, 0, "Delayed ack mode (default = 0)"); int c4iw_max_read_depth = 8; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RWTUN, &c4iw_max_read_depth, 0, "Per-connection max ORD/IRD (default = 8)"); static int enable_tcp_timestamps; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RWTUN, &enable_tcp_timestamps, 0, "Enable tcp timestamps (default = 0)"); static int enable_tcp_sack; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RWTUN, &enable_tcp_sack, 0, "Enable tcp SACK (default = 0)"); static int enable_tcp_window_scaling = 1; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RWTUN, &enable_tcp_window_scaling, 0, "Enable tcp window scaling (default = 1)"); int c4iw_debug = 0; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RWTUN, &c4iw_debug, 0, "Enable debug logging (default = 0)"); static int peer2peer = 1; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RWTUN, &peer2peer, 0, "Support peer2peer ULPs (default = 1)"); static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RWTUN, &p2p_type, 0, "RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)"); static int ep_timeout_secs = 60; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RWTUN, &ep_timeout_secs, 0, "CM Endpoint operation timeout in seconds (default = 60)"); static int mpa_rev = 1; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RWTUN, &mpa_rev, 0, "MPA Revision, 0 supports amso1100, 1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)"); static int markers_enabled; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RWTUN, &markers_enabled, 0, "Enable MPA MARKERS (default(0) = disabled)"); static int crc_enabled = 1; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RWTUN, &crc_enabled, 0, "Enable MPA CRC (default(1) = enabled)"); static int rcv_win = 256 * 1024; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RWTUN, &rcv_win, 0, "TCP receive window in bytes (default = 256KB)"); static int snd_win = 128 * 1024; SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RWTUN, &snd_win, 0, "TCP send window in bytes (default = 128KB)"); static void start_ep_timer(struct c4iw_ep *ep) { if (timer_pending(&ep->timer)) { CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep); printk(KERN_ERR "%s timer already started! ep %p\n", __func__, ep); return; } clear_bit(TIMEOUT, &ep->com.flags); c4iw_get_ep(&ep->com); ep->timer.expires = jiffies + ep_timeout_secs * HZ; ep->timer.data = (unsigned long)ep; ep->timer.function = ep_timeout; add_timer(&ep->timer); } static int stop_ep_timer(struct c4iw_ep *ep) { del_timer_sync(&ep->timer); if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { c4iw_put_ep(&ep->com); return 0; } return 1; } static void * alloc_ep(int size, gfp_t gfp) { struct c4iw_ep_common *epc; epc = kzalloc(size, gfp); if (epc == NULL) return (NULL); kref_init(&epc->kref); mutex_init(&epc->mutex); c4iw_init_wr_wait(&epc->wr_wait); return (epc); } void _c4iw_free_ep(struct kref *kref) { struct c4iw_ep *ep; struct c4iw_ep_common *epc; ep = container_of(kref, struct c4iw_ep, com.kref); epc = &ep->com; KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list", __func__, epc)); if (test_bit(QP_REFERENCED, &ep->com.flags)) deref_qp(ep); CTR4(KTR_IW_CXGBE, "%s: ep %p, history 0x%lx, flags 0x%lx", __func__, ep, epc->history, epc->flags); kfree(ep); } static void release_ep_resources(struct c4iw_ep *ep) { CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep); set_bit(RELEASE_RESOURCES, &ep->com.flags); c4iw_put_ep(&ep->com); CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep); } static int send_mpa_req(struct c4iw_ep *ep) { int mpalen; struct mpa_message *mpa; struct mpa_v2_conn_params mpa_v2_params; struct mbuf *m; char mpa_rev_to_use = mpa_rev; int err = 0; if (ep->retry_with_mpa_v1) mpa_rev_to_use = 1; mpalen = sizeof(*mpa) + ep->plen; if (mpa_rev_to_use == 2) mpalen += sizeof(struct mpa_v2_conn_params); mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); if (mpa == NULL) { err = -ENOMEM; CTR3(KTR_IW_CXGBE, "%s:smr1 ep: %p , error: %d", __func__, ep, err); goto err; } memset(mpa, 0, mpalen); memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); mpa->flags = (crc_enabled ? MPA_CRC : 0) | (markers_enabled ? MPA_MARKERS : 0) | (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0); mpa->private_data_size = htons(ep->plen); mpa->revision = mpa_rev_to_use; if (mpa_rev_to_use == 1) { ep->tried_with_mpa_v1 = 1; ep->retry_with_mpa_v1 = 0; } if (mpa_rev_to_use == 2) { mpa->private_data_size = htons(ntohs(mpa->private_data_size) + sizeof(struct mpa_v2_conn_params)); mpa_v2_params.ird = htons((u16)ep->ird); mpa_v2_params.ord = htons((u16)ep->ord); if (peer2peer) { mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { mpa_v2_params.ord |= htons(MPA_V2_RDMA_WRITE_RTR); } else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { mpa_v2_params.ord |= htons(MPA_V2_RDMA_READ_RTR); } } memcpy(mpa->private_data, &mpa_v2_params, sizeof(struct mpa_v2_conn_params)); if (ep->plen) { memcpy(mpa->private_data + sizeof(struct mpa_v2_conn_params), ep->mpa_pkt + sizeof(*mpa), ep->plen); } } else { if (ep->plen) memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen); CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep); } m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); if (m == NULL) { err = -ENOMEM; CTR3(KTR_IW_CXGBE, "%s:smr2 ep: %p , error: %d", __func__, ep, err); free(mpa, M_CXGBE); goto err; } m_copyback(m, 0, mpalen, (void *)mpa); free(mpa, M_CXGBE); err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread); if (err) { CTR3(KTR_IW_CXGBE, "%s:smr3 ep: %p , error: %d", __func__, ep, err); goto err; } START_EP_TIMER(ep); ep->com.state = MPA_REQ_SENT; ep->mpa_attr.initiator = 1; CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); return 0; err: connect_reply_upcall(ep, err); CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); return err; } static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen) { int mpalen ; struct mpa_message *mpa; struct mpa_v2_conn_params mpa_v2_params; struct mbuf *m; int err; CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid, ep->plen); mpalen = sizeof(*mpa) + plen; if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { mpalen += sizeof(struct mpa_v2_conn_params); CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep, ep->mpa_attr.version, mpalen); } mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); if (mpa == NULL) return (-ENOMEM); memset(mpa, 0, mpalen); memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); mpa->flags = MPA_REJECT; mpa->revision = mpa_rev; mpa->private_data_size = htons(plen); if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { mpa->flags |= MPA_ENHANCED_RDMA_CONN; mpa->private_data_size = htons(ntohs(mpa->private_data_size) + sizeof(struct mpa_v2_conn_params)); mpa_v2_params.ird = htons(((u16)ep->ird) | (peer2peer ? MPA_V2_PEER2PEER_MODEL : 0)); mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ? (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE ? MPA_V2_RDMA_WRITE_RTR : p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ ? MPA_V2_RDMA_READ_RTR : 0) : 0)); memcpy(mpa->private_data, &mpa_v2_params, sizeof(struct mpa_v2_conn_params)); if (ep->plen) memcpy(mpa->private_data + sizeof(struct mpa_v2_conn_params), pdata, plen); CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep, mpa_v2_params.ird, mpa_v2_params.ord, ep->plen); } else if (plen) memcpy(mpa->private_data, pdata, plen); m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); if (m == NULL) { free(mpa, M_CXGBE); return (-ENOMEM); } m_copyback(m, 0, mpalen, (void *)mpa); free(mpa, M_CXGBE); err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread); if (!err) ep->snd_seq += mpalen; CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err); return err; } static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen) { int mpalen; struct mpa_message *mpa; struct mbuf *m; struct mpa_v2_conn_params mpa_v2_params; int err; CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep); mpalen = sizeof(*mpa) + plen; if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep, ep->mpa_attr.version); mpalen += sizeof(struct mpa_v2_conn_params); } mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); if (mpa == NULL) return (-ENOMEM); memset(mpa, 0, sizeof(*mpa)); memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) | (markers_enabled ? MPA_MARKERS : 0); mpa->revision = ep->mpa_attr.version; mpa->private_data_size = htons(plen); if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { mpa->flags |= MPA_ENHANCED_RDMA_CONN; mpa->private_data_size += htons(sizeof(struct mpa_v2_conn_params)); mpa_v2_params.ird = htons((u16)ep->ird); mpa_v2_params.ord = htons((u16)ep->ord); CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep, ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord); if (peer2peer && (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED)) { mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { mpa_v2_params.ord |= htons(MPA_V2_RDMA_WRITE_RTR); CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d", __func__, ep, p2p_type, mpa_v2_params.ird, mpa_v2_params.ord); } else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { mpa_v2_params.ord |= htons(MPA_V2_RDMA_READ_RTR); CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d", __func__, ep, p2p_type, mpa_v2_params.ird, mpa_v2_params.ord); } } memcpy(mpa->private_data, &mpa_v2_params, sizeof(struct mpa_v2_conn_params)); if (ep->plen) memcpy(mpa->private_data + sizeof(struct mpa_v2_conn_params), pdata, plen); } else if (plen) memcpy(mpa->private_data, pdata, plen); m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); if (m == NULL) { free(mpa, M_CXGBE); return (-ENOMEM); } m_copyback(m, 0, mpalen, (void *)mpa); free(mpa, M_CXGBE); ep->com.state = MPA_REP_SENT; ep->snd_seq += mpalen; err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread); CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err); return err; } static void close_complete_upcall(struct c4iw_ep *ep, int status) { struct iw_cm_event event; CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep); memset(&event, 0, sizeof(event)); event.event = IW_CM_EVENT_CLOSE; event.status = status; if (ep->com.cm_id) { CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep); ep->com.cm_id->event_handler(ep->com.cm_id, &event); deref_cm_id(&ep->com); set_bit(CLOSE_UPCALL, &ep->com.history); } CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep); } static int send_abort(struct c4iw_ep *ep) { struct socket *so = ep->com.so; struct sockopt sopt; int rc; struct linger l; CTR5(KTR_IW_CXGBE, "%s ep %p so %p state %s tid %d", __func__, ep, so, states[ep->com.state], ep->hwtid); l.l_onoff = 1; l.l_linger = 0; /* linger_time of 0 forces RST to be sent */ sopt.sopt_dir = SOPT_SET; sopt.sopt_level = SOL_SOCKET; sopt.sopt_name = SO_LINGER; sopt.sopt_val = (caddr_t)&l; sopt.sopt_valsize = sizeof l; sopt.sopt_td = NULL; rc = sosetopt(so, &sopt); if (rc != 0) { log(LOG_ERR, "%s: sosetopt(%p, linger = 0) failed with %d.\n", __func__, so, rc); } uninit_iwarp_socket(so); soclose(so); set_bit(ABORT_CONN, &ep->com.history); /* * TBD: iw_cxgbe driver should receive ABORT reply for every ABORT * request it has sent. But the current TOE driver is not propagating * this ABORT reply event (via do_abort_rpl) to iw_cxgbe. So as a work- * around de-refererece 'ep' here instead of doing it in abort_rpl() * handler(not yet implemented) of iw_cxgbe driver. */ release_ep_resources(ep); + ep->com.state = DEAD; return (0); } static void peer_close_upcall(struct c4iw_ep *ep) { struct iw_cm_event event; CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep); memset(&event, 0, sizeof(event)); event.event = IW_CM_EVENT_DISCONNECT; if (ep->com.cm_id) { CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep); ep->com.cm_id->event_handler(ep->com.cm_id, &event); set_bit(DISCONN_UPCALL, &ep->com.history); } CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep); } static void peer_abort_upcall(struct c4iw_ep *ep) { struct iw_cm_event event; CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep); memset(&event, 0, sizeof(event)); event.event = IW_CM_EVENT_CLOSE; event.status = -ECONNRESET; if (ep->com.cm_id) { CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep); ep->com.cm_id->event_handler(ep->com.cm_id, &event); deref_cm_id(&ep->com); set_bit(ABORT_UPCALL, &ep->com.history); } CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep); } static void connect_reply_upcall(struct c4iw_ep *ep, int status) { struct iw_cm_event event; CTR3(KTR_IW_CXGBE, "%s:cruB %p, status: %d", __func__, ep, status); memset(&event, 0, sizeof(event)); event.event = IW_CM_EVENT_CONNECT_REPLY; event.status = ((status == -ECONNABORTED) || (status == -EPIPE)) ? -ECONNRESET : status; event.local_addr = ep->com.local_addr; event.remote_addr = ep->com.remote_addr; if ((status == 0) || (status == -ECONNREFUSED)) { if (!ep->tried_with_mpa_v1) { CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep); /* this means MPA_v2 is used */ event.ord = ep->ird; event.ird = ep->ord; event.private_data_len = ep->plen - sizeof(struct mpa_v2_conn_params); event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + sizeof(struct mpa_v2_conn_params); } else { CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep); /* this means MPA_v1 is used */ event.ord = c4iw_max_read_depth; event.ird = c4iw_max_read_depth; event.private_data_len = ep->plen; event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); } } if (ep->com.cm_id) { CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep); set_bit(CONN_RPL_UPCALL, &ep->com.history); ep->com.cm_id->event_handler(ep->com.cm_id, &event); } if(status == -ECONNABORTED) { CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status); return; } if (status < 0) { CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status); deref_cm_id(&ep->com); } CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep); } static int connect_request_upcall(struct c4iw_ep *ep) { struct iw_cm_event event; int ret; CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep, ep->tried_with_mpa_v1); memset(&event, 0, sizeof(event)); event.event = IW_CM_EVENT_CONNECT_REQUEST; event.local_addr = ep->com.local_addr; event.remote_addr = ep->com.remote_addr; event.provider_data = ep; if (!ep->tried_with_mpa_v1) { /* this means MPA_v2 is used */ event.ord = ep->ord; event.ird = ep->ird; event.private_data_len = ep->plen - sizeof(struct mpa_v2_conn_params); event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + sizeof(struct mpa_v2_conn_params); } else { /* this means MPA_v1 is used. Send max supported */ event.ord = c4iw_max_read_depth; event.ird = c4iw_max_read_depth; event.private_data_len = ep->plen; event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); } c4iw_get_ep(&ep->com); ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, &event); if(ret) { CTR3(KTR_IW_CXGBE, "%s: ep %p, Failure while notifying event to" " IWCM, err:%d", __func__, ep, ret); c4iw_put_ep(&ep->com); } else /* Dereference parent_ep only in success case. * In case of failure, parent_ep is dereferenced by the caller * of process_mpa_request(). */ c4iw_put_ep(&ep->parent_ep->com); set_bit(CONNREQ_UPCALL, &ep->com.history); return ret; } static void established_upcall(struct c4iw_ep *ep) { struct iw_cm_event event; CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep); memset(&event, 0, sizeof(event)); event.event = IW_CM_EVENT_ESTABLISHED; event.ird = ep->ord; event.ord = ep->ird; if (ep->com.cm_id) { CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep); ep->com.cm_id->event_handler(ep->com.cm_id, &event); set_bit(ESTAB_UPCALL, &ep->com.history); } CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep); } #define RELAXED_IRD_NEGOTIATION 1 /* * process_mpa_reply - process streaming mode MPA reply * * Returns: * * 0 upon success indicating a connect request was delivered to the ULP * or the mpa request is incomplete but valid so far. * * 1 if a failure requires the caller to close the connection. * * 2 if a failure requires the caller to abort the connection. */ static int process_mpa_reply(struct c4iw_ep *ep) { struct mpa_message *mpa; struct mpa_v2_conn_params *mpa_v2_params; u16 plen; u16 resp_ird, resp_ord; u8 rtr_mismatch = 0, insuff_ird = 0; struct c4iw_qp_attributes attrs = {0}; enum c4iw_qp_attr_mask mask; int err; struct mbuf *top, *m; int flags = MSG_DONTWAIT; struct uio uio; int disconnect = 0; CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep); /* * Stop mpa timer. If it expired, then * we ignore the MPA reply. process_timeout() * will abort the connection. */ if (STOP_EP_TIMER(ep)) return 0; uio.uio_resid = 1000000; uio.uio_td = ep->com.thread; err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags); if (err) { if (err == EWOULDBLOCK) { CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep); START_EP_TIMER(ep); return 0; } err = -err; CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep); goto err; } if (ep->com.so->so_rcv.sb_mb) { CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep); printf("%s data after soreceive called! so %p sb_mb %p top %p\n", __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top); } m = top; do { CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep); /* * If we get more than the supported amount of private data * then we must fail this connection. */ if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) { CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep, ep->mpa_pkt_len + m->m_len); err = (-EINVAL); goto err_stop_timer; } /* * copy the new data into our accumulation buffer. */ m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len])); ep->mpa_pkt_len += m->m_len; if (!m->m_next) m = m->m_nextpkt; else m = m->m_next; } while (m); m_freem(top); /* * if we don't even have the mpa message, then bail. */ if (ep->mpa_pkt_len < sizeof(*mpa)) { return 0; } mpa = (struct mpa_message *) ep->mpa_pkt; /* Validate MPA header. */ if (mpa->revision > mpa_rev) { CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep, mpa->revision, mpa_rev); printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, " " Received = %d\n", __func__, mpa_rev, mpa->revision); err = -EPROTO; goto err_stop_timer; } if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep); err = -EPROTO; goto err_stop_timer; } plen = ntohs(mpa->private_data_size); /* * Fail if there's too much private data. */ if (plen > MPA_MAX_PRIVATE_DATA) { CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep); err = -EPROTO; goto err_stop_timer; } /* * If plen does not account for pkt size */ if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep); STOP_EP_TIMER(ep); err = -EPROTO; goto err_stop_timer; } ep->plen = (u8) plen; /* * If we don't have all the pdata yet, then bail. * We'll continue process when more data arrives. */ if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) { CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep); return 0; } if (mpa->flags & MPA_REJECT) { CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep); err = -ECONNREFUSED; goto err_stop_timer; } /* * If we get here we have accumulated the entire mpa * start reply message including private data. And * the MPA header is valid. */ ep->com.state = FPDU_MODE; ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; ep->mpa_attr.recv_marker_enabled = markers_enabled; ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; ep->mpa_attr.version = mpa->revision; ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; if (mpa->revision == 2) { CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep); ep->mpa_attr.enhanced_rdma_conn = mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; if (ep->mpa_attr.enhanced_rdma_conn) { CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep); mpa_v2_params = (struct mpa_v2_conn_params *) (ep->mpa_pkt + sizeof(*mpa)); resp_ird = ntohs(mpa_v2_params->ird) & MPA_V2_IRD_ORD_MASK; resp_ord = ntohs(mpa_v2_params->ord) & MPA_V2_IRD_ORD_MASK; /* * This is a double-check. Ideally, below checks are * not required since ird/ord stuff has been taken * care of in c4iw_accept_cr */ if (ep->ird < resp_ord) { if (RELAXED_IRD_NEGOTIATION && resp_ord <= ep->com.dev->rdev.adap->params.max_ordird_qp) ep->ird = resp_ord; else insuff_ird = 1; } else if (ep->ird > resp_ord) { ep->ird = resp_ord; } if (ep->ord > resp_ird) { if (RELAXED_IRD_NEGOTIATION) ep->ord = resp_ird; else insuff_ird = 1; } if (insuff_ird) { err = -ENOMEM; ep->ird = resp_ord; ep->ord = resp_ird; } if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) { CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep); if (ntohs(mpa_v2_params->ord) & MPA_V2_RDMA_WRITE_RTR) { CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep); ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_RDMA_WRITE; } else if (ntohs(mpa_v2_params->ord) & MPA_V2_RDMA_READ_RTR) { CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep); ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; } } } } else { CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep); if (mpa->revision == 1) { CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep); if (peer2peer) { CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep); ep->mpa_attr.p2p_type = p2p_type; } } } if (set_tcpinfo(ep)) { CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep); printf("%s set_tcpinfo error\n", __func__); err = -ECONNRESET; goto err; } CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, " "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__, ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, ep->mpa_attr.p2p_type); /* * If responder's RTR does not match with that of initiator, assign * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not * generated when moving QP to RTS state. * A TERM message will be sent after QP has moved to RTS state */ if ((ep->mpa_attr.version == 2) && peer2peer && (ep->mpa_attr.p2p_type != p2p_type)) { CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep); ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; rtr_mismatch = 1; } //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; attrs.mpa_attr = ep->mpa_attr; attrs.max_ird = ep->ird; attrs.max_ord = ep->ord; attrs.llp_stream_handle = ep; attrs.next_state = C4IW_QP_STATE_RTS; mask = C4IW_QP_ATTR_NEXT_STATE | C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; /* bind QP and TID with INIT_WR */ err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); if (err) { CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep); goto err; } /* * If responder's RTR requirement did not match with what initiator * supports, generate TERM message */ if (rtr_mismatch) { CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep); printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__); attrs.layer_etype = LAYER_MPA | DDP_LLP; attrs.ecode = MPA_NOMATCH_RTR; attrs.next_state = C4IW_QP_STATE_TERMINATE; attrs.send_term = 1; err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); err = -ENOMEM; disconnect = 1; goto out; } /* * Generate TERM if initiator IRD is not sufficient for responder * provided ORD. Currently, we do the same behaviour even when * responder provided IRD is also not sufficient as regards to * initiator ORD. */ if (insuff_ird) { CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep); printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n", __func__); attrs.layer_etype = LAYER_MPA | DDP_LLP; attrs.ecode = MPA_INSUFF_IRD; attrs.next_state = C4IW_QP_STATE_TERMINATE; attrs.send_term = 1; err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); err = -ENOMEM; disconnect = 1; goto out; } goto out; err_stop_timer: STOP_EP_TIMER(ep); err: disconnect = 2; out: connect_reply_upcall(ep, err); CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep); return disconnect; } /* * process_mpa_request - process streaming mode MPA request * * Returns: * * 0 upon success indicating a connect request was delivered to the ULP * or the mpa request is incomplete but valid so far. * * 1 if a failure requires the caller to close the connection. * * 2 if a failure requires the caller to abort the connection. */ static int process_mpa_request(struct c4iw_ep *ep) { struct mpa_message *mpa; struct mpa_v2_conn_params *mpa_v2_params; u16 plen; int flags = MSG_DONTWAIT; int rc; struct iovec iov; struct uio uio; enum c4iw_ep_state state = ep->com.state; CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]); if (state != MPA_REQ_WAIT) return 0; iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len]; iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */ rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags); if (rc == EAGAIN) return 0; else if (rc) goto err_stop_timer; KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data", __func__, ep->com.so)); ep->mpa_pkt_len += uio.uio_offset; /* * If we get more than the supported amount of private data then we must * fail this connection. XXX: check so_rcv->sb_cc, or peek with another * soreceive, or increase the size of mpa_pkt by 1 and abort if the last * byte is filled by the soreceive above. */ /* Don't even have the MPA message. Wait for more data to arrive. */ if (ep->mpa_pkt_len < sizeof(*mpa)) return 0; mpa = (struct mpa_message *) ep->mpa_pkt; /* * Validate MPA Header. */ if (mpa->revision > mpa_rev) { log(LOG_ERR, "%s: MPA version mismatch. Local = %d," " Received = %d\n", __func__, mpa_rev, mpa->revision); goto err_stop_timer; } if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) goto err_stop_timer; /* * Fail if there's too much private data. */ plen = ntohs(mpa->private_data_size); if (plen > MPA_MAX_PRIVATE_DATA) goto err_stop_timer; /* * If plen does not account for pkt size */ if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) goto err_stop_timer; ep->plen = (u8) plen; /* * If we don't have all the pdata yet, then bail. */ if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) return 0; /* * If we get here we have accumulated the entire mpa * start reply message including private data. */ ep->mpa_attr.initiator = 0; ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; ep->mpa_attr.recv_marker_enabled = markers_enabled; ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; ep->mpa_attr.version = mpa->revision; if (mpa->revision == 1) ep->tried_with_mpa_v1 = 1; ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; if (mpa->revision == 2) { ep->mpa_attr.enhanced_rdma_conn = mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; if (ep->mpa_attr.enhanced_rdma_conn) { mpa_v2_params = (struct mpa_v2_conn_params *) (ep->mpa_pkt + sizeof(*mpa)); ep->ird = ntohs(mpa_v2_params->ird) & MPA_V2_IRD_ORD_MASK; ep->ird = min_t(u32, ep->ird, cur_max_read_depth(ep->com.dev)); ep->ord = ntohs(mpa_v2_params->ord) & MPA_V2_IRD_ORD_MASK; ep->ord = min_t(u32, ep->ord, cur_max_read_depth(ep->com.dev)); CTR3(KTR_IW_CXGBE, "%s initiator ird %u ord %u\n", __func__, ep->ird, ep->ord); if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) if (peer2peer) { if (ntohs(mpa_v2_params->ord) & MPA_V2_RDMA_WRITE_RTR) ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_RDMA_WRITE; else if (ntohs(mpa_v2_params->ord) & MPA_V2_RDMA_READ_RTR) ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; } } } else if (mpa->revision == 1 && peer2peer) ep->mpa_attr.p2p_type = p2p_type; if (set_tcpinfo(ep)) goto err_stop_timer; CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, " "xmit_marker_enabled = %d, version = %d", __func__, ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version); ep->com.state = MPA_REQ_RCVD; STOP_EP_TIMER(ep); /* drive upcall */ if (ep->parent_ep->com.state != DEAD) if (connect_request_upcall(ep)) goto err_out; return 0; err_stop_timer: STOP_EP_TIMER(ep); err_out: return 2; } /* * Upcall from the adapter indicating data has been transmitted. * For us its just the single MPA request or reply. We can now free * the skb holding the mpa message. */ int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) { int err; struct c4iw_ep *ep = to_ep(cm_id); int abort = 0; mutex_lock(&ep->com.mutex); CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep); if ((ep->com.state == DEAD) || (ep->com.state != MPA_REQ_RCVD)) { CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep); mutex_unlock(&ep->com.mutex); c4iw_put_ep(&ep->com); return -ECONNRESET; } set_bit(ULP_REJECT, &ep->com.history); if (mpa_rev == 0) { CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep); abort = 1; } else { CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep); abort = send_mpa_reject(ep, pdata, pdata_len); } STOP_EP_TIMER(ep); err = c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL); mutex_unlock(&ep->com.mutex); c4iw_put_ep(&ep->com); CTR3(KTR_IW_CXGBE, "%s:crc4 %p, err: %d", __func__, ep, err); return 0; } int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) { int err; struct c4iw_qp_attributes attrs = {0}; enum c4iw_qp_attr_mask mask; struct c4iw_ep *ep = to_ep(cm_id); struct c4iw_dev *h = to_c4iw_dev(cm_id->device); struct c4iw_qp *qp = get_qhp(h, conn_param->qpn); int abort = 0; mutex_lock(&ep->com.mutex); CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep); if ((ep->com.state == DEAD) || (ep->com.state != MPA_REQ_RCVD)) { CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep); err = -ECONNRESET; goto err_out; } BUG_ON(!qp); set_bit(ULP_ACCEPT, &ep->com.history); if ((conn_param->ord > c4iw_max_read_depth) || (conn_param->ird > c4iw_max_read_depth)) { CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep); err = -EINVAL; goto err_abort; } if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep); if (conn_param->ord > ep->ird) { if (RELAXED_IRD_NEGOTIATION) { conn_param->ord = ep->ird; } else { ep->ird = conn_param->ird; ep->ord = conn_param->ord; send_mpa_reject(ep, conn_param->private_data, conn_param->private_data_len); err = -ENOMEM; goto err_abort; } } if (conn_param->ird < ep->ord) { if (RELAXED_IRD_NEGOTIATION && ep->ord <= h->rdev.adap->params.max_ordird_qp) { conn_param->ird = ep->ord; } else { err = -ENOMEM; goto err_abort; } } } ep->ird = conn_param->ird; ep->ord = conn_param->ord; if (ep->mpa_attr.version == 1) { if (peer2peer && ep->ird == 0) ep->ird = 1; } else { if (peer2peer && (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) && (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0) ep->ird = 1; } CTR4(KTR_IW_CXGBE, "%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord); ep->com.cm_id = cm_id; ref_cm_id(&ep->com); ep->com.qp = qp; ref_qp(ep); //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; /* bind QP to EP and move to RTS */ attrs.mpa_attr = ep->mpa_attr; attrs.max_ird = ep->ird; attrs.max_ord = ep->ord; attrs.llp_stream_handle = ep; attrs.next_state = C4IW_QP_STATE_RTS; /* bind QP and TID with INIT_WR */ mask = C4IW_QP_ATTR_NEXT_STATE | C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); if (err) { CTR3(KTR_IW_CXGBE, "%s:caca %p, err: %d", __func__, ep, err); goto err_defef_cm_id; } err = send_mpa_reply(ep, conn_param->private_data, conn_param->private_data_len); if (err) { CTR3(KTR_IW_CXGBE, "%s:cacb %p, err: %d", __func__, ep, err); goto err_defef_cm_id; } ep->com.state = FPDU_MODE; established_upcall(ep); mutex_unlock(&ep->com.mutex); c4iw_put_ep(&ep->com); CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep); return 0; err_defef_cm_id: deref_cm_id(&ep->com); err_abort: abort = 1; err_out: if (abort) c4iw_ep_disconnect(ep, 1, GFP_KERNEL); mutex_unlock(&ep->com.mutex); c4iw_put_ep(&ep->com); CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep); return err; } static int c4iw_sock_create(struct sockaddr_storage *laddr, struct socket **so) { int ret; int size; struct socket *sock = NULL; ret = sock_create_kern(laddr->ss_family, SOCK_STREAM, IPPROTO_TCP, &sock); if (ret) { CTR2(KTR_IW_CXGBE, "%s:Failed to create TCP socket. err %d", __func__, ret); return ret; } ret = sobind(sock, (struct sockaddr *)laddr, curthread); if (ret) { CTR2(KTR_IW_CXGBE, "%s:Failed to bind socket. err %p", __func__, ret); sock_release(sock); return ret; } size = laddr->ss_family == AF_INET6 ? sizeof(struct sockaddr_in6) : sizeof(struct sockaddr_in); ret = sock_getname(sock, (struct sockaddr *)laddr, &size, 0); if (ret) { CTR2(KTR_IW_CXGBE, "%s:sock_getname failed. err %p", __func__, ret); sock_release(sock); return ret; } *so = sock; return 0; } int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) { int err = 0; struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); struct c4iw_ep *ep = NULL; struct ifnet *nh_ifp; /* Logical egress interface */ #ifdef VIMAGE struct rdma_cm_id *rdma_id = (struct rdma_cm_id*)cm_id->context; struct vnet *vnet = rdma_id->route.addr.dev_addr.net; #endif CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id); if ((conn_param->ord > c4iw_max_read_depth) || (conn_param->ird > c4iw_max_read_depth)) { CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id); err = -EINVAL; goto out; } ep = alloc_ep(sizeof(*ep), GFP_KERNEL); cm_id->provider_data = ep; init_timer(&ep->timer); ep->plen = conn_param->private_data_len; if (ep->plen) { CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep); memcpy(ep->mpa_pkt + sizeof(struct mpa_message), conn_param->private_data, ep->plen); } ep->ird = conn_param->ird; ep->ord = conn_param->ord; if (peer2peer && ep->ord == 0) { CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep); ep->ord = 1; } ep->com.dev = dev; ep->com.cm_id = cm_id; ref_cm_id(&ep->com); ep->com.qp = get_qhp(dev, conn_param->qpn); if (!ep->com.qp) { CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep); err = -EINVAL; goto fail; } ref_qp(ep); ep->com.thread = curthread; CURVNET_SET(vnet); err = get_ifnet_from_raddr(&cm_id->remote_addr, &nh_ifp); CURVNET_RESTORE(); if (err) { CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep); printk(KERN_ERR MOD "%s - cannot find route.\n", __func__); err = EHOSTUNREACH; return err; } if (!(nh_ifp->if_capenable & IFCAP_TOE) || TOEDEV(nh_ifp) == NULL) { err = -ENOPROTOOPT; goto fail; } ep->com.state = CONNECTING; ep->tos = 0; ep->com.local_addr = cm_id->local_addr; ep->com.remote_addr = cm_id->remote_addr; err = c4iw_sock_create(&cm_id->local_addr, &ep->com.so); if (err) goto fail; setiwsockopt(ep->com.so); init_iwarp_socket(ep->com.so, &ep->com); err = -soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr, ep->com.thread); if (err) goto fail_free_so; CTR2(KTR_IW_CXGBE, "%s:ccE, ep %p", __func__, ep); return 0; fail_free_so: uninit_iwarp_socket(ep->com.so); ep->com.state = DEAD; sock_release(ep->com.so); fail: deref_cm_id(&ep->com); c4iw_put_ep(&ep->com); ep = NULL; out: CTR2(KTR_IW_CXGBE, "%s:ccE Error %d", __func__, err); return err; } /* * iwcm->create_listen. Returns -errno on failure. */ int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) { struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); struct c4iw_listen_ep *lep = NULL; struct listen_port_info *port_info = NULL; int rc = 0; CTR3(KTR_IW_CXGBE, "%s: cm_id %p, backlog %s", __func__, cm_id, backlog); lep = alloc_ep(sizeof(*lep), GFP_KERNEL); lep->com.cm_id = cm_id; ref_cm_id(&lep->com); lep->com.dev = dev; lep->backlog = backlog; lep->com.local_addr = cm_id->local_addr; lep->com.thread = curthread; cm_id->provider_data = lep; lep->com.state = LISTEN; /* In case of INDADDR_ANY, ibcore creates cmid for each device and * invokes iw_cxgbe listener callbacks assuming that iw_cxgbe creates * HW listeners for each device seperately. But toecore expects single * solisten() call with INADDR_ANY address to create HW listeners on * all devices for a given port number. So iw_cxgbe driver calls * solisten() only once for INADDR_ANY(usually done at first time * listener callback from ibcore). And all the subsequent INADDR_ANY * listener callbacks from ibcore(for the same port address) do not * invoke solisten() as first listener callback has already created * listeners for all other devices(via solisten). */ if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) { port_info = add_ep_to_listenlist(lep); /* skip solisten() if refcnt > 1, as the listeners were * alredy created by 'Master lep' */ if (port_info->refcnt > 1) { /* As there will be only one listener socket for a TCP * port, copy Master lep's socket pointer to other lep's * that are belonging to same TCP port. */ struct c4iw_listen_ep *head_lep = container_of(port_info->lep_list.next, struct c4iw_listen_ep, listen_ep_list); lep->com.so = head_lep->com.so; goto out; } } rc = c4iw_sock_create(&cm_id->local_addr, &lep->com.so); if (rc) { CTR2(KTR_IW_CXGBE, "%s:Failed to create socket. err %d", __func__, rc); goto fail; } rc = solisten(lep->com.so, backlog, curthread); if (rc) { CTR3(KTR_IW_CXGBE, "%s:Failed to listen on sock:%p. err %d", __func__, lep->com.so, rc); goto fail_free_so; } init_iwarp_socket(lep->com.so, &lep->com); out: return 0; fail_free_so: sock_release(lep->com.so); fail: if (port_info) rem_ep_from_listenlist(lep); deref_cm_id(&lep->com); c4iw_put_ep(&lep->com); return rc; } int c4iw_destroy_listen(struct iw_cm_id *cm_id) { struct c4iw_listen_ep *lep = to_listen_ep(cm_id); mutex_lock(&lep->com.mutex); CTR3(KTR_IW_CXGBE, "%s: cm_id %p, state %s", __func__, cm_id, states[lep->com.state]); lep->com.state = DEAD; if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) { /* if no refcount then close listen socket */ if (!rem_ep_from_listenlist(lep)) close_socket(lep->com.so); } else close_socket(lep->com.so); deref_cm_id(&lep->com); mutex_unlock(&lep->com.mutex); c4iw_put_ep(&lep->com); return 0; } int __c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) { int ret; mutex_lock(&ep->com.mutex); ret = c4iw_ep_disconnect(ep, abrupt, gfp); mutex_unlock(&ep->com.mutex); return ret; } int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) { int ret = 0; int close = 0; int fatal = 0; struct c4iw_rdev *rdev; CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep); rdev = &ep->com.dev->rdev; if (c4iw_fatal_error(rdev)) { CTR2(KTR_IW_CXGBE, "%s:ced1 %p", __func__, ep); fatal = 1; close_complete_upcall(ep, -ECONNRESET); send_abort(ep); ep->com.state = DEAD; } CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep, states[ep->com.state]); /* * Ref the ep here in case we have fatal errors causing the * ep to be released and freed. */ c4iw_get_ep(&ep->com); switch (ep->com.state) { case MPA_REQ_WAIT: case MPA_REQ_SENT: case MPA_REQ_RCVD: case MPA_REP_SENT: case FPDU_MODE: close = 1; if (abrupt) ep->com.state = ABORTING; else { ep->com.state = CLOSING; START_EP_TIMER(ep); } set_bit(CLOSE_SENT, &ep->com.flags); break; case CLOSING: if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) { close = 1; if (abrupt) { STOP_EP_TIMER(ep); ep->com.state = ABORTING; } else ep->com.state = MORIBUND; } break; case MORIBUND: case ABORTING: case DEAD: CTR3(KTR_IW_CXGBE, "%s ignoring disconnect ep %p state %u", __func__, ep, ep->com.state); break; default: BUG(); break; } if (close) { CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep); if (abrupt) { CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep); set_bit(EP_DISC_ABORT, &ep->com.history); close_complete_upcall(ep, -ECONNRESET); ret = send_abort(ep); if (ret) fatal = 1; } else { CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep); set_bit(EP_DISC_CLOSE, &ep->com.history); if (!ep->parent_ep) ep->com.state = MORIBUND; CURVNET_SET(ep->com.so->so_vnet); sodisconnect(ep->com.so); CURVNET_RESTORE(); } } if (fatal) { set_bit(EP_DISC_FAIL, &ep->com.history); if (!abrupt) { STOP_EP_TIMER(ep); close_complete_upcall(ep, -EIO); } if (ep->com.qp) { struct c4iw_qp_attributes attrs = {0}; attrs.next_state = C4IW_QP_STATE_ERROR; ret = c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); if (ret) { CTR2(KTR_IW_CXGBE, "%s:ced7 %p", __func__, ep); printf("%s - qp <- error failed!\n", __func__); } } release_ep_resources(ep); ep->com.state = DEAD; CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep); } c4iw_put_ep(&ep->com); CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep); return ret; } #ifdef C4IW_EP_REDIRECT int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new, struct l2t_entry *l2t) { struct c4iw_ep *ep = ctx; if (ep->dst != old) return 0; PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new, l2t); dst_hold(new); cxgb4_l2t_release(ep->l2t); ep->l2t = l2t; dst_release(old); ep->dst = new; return 1; } #endif static void ep_timeout(unsigned long arg) { struct c4iw_ep *ep = (struct c4iw_ep *)arg; if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { /* * Only insert if it is not already on the list. */ if (!(ep->com.ep_events & C4IW_EVENT_TIMEOUT)) { CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep); add_ep_to_req_list(ep, C4IW_EVENT_TIMEOUT); } } } static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl) { uint64_t val = be64toh(*rpl); int ret; struct c4iw_wr_wait *wr_waitp; ret = (int)((val >> 8) & 0xff); wr_waitp = (struct c4iw_wr_wait *)rpl[1]; CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret); if (wr_waitp) c4iw_wake_up(wr_waitp, ret ? -ret : 0); return (0); } static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl) { struct cqe_list_entry *cle; unsigned long flag; cle = malloc(sizeof(*cle), M_CXGBE, M_NOWAIT); cle->rhp = sc->iwarp_softc; cle->err_cqe = *(const struct t4_cqe *)(&rpl[0]); spin_lock_irqsave(&err_cqe_lock, flag); list_add_tail(&cle->entry, &err_cqe_list); queue_work(c4iw_taskq, &c4iw_task); spin_unlock_irqrestore(&err_cqe_lock, flag); return (0); } static int process_terminate(struct c4iw_ep *ep) { struct c4iw_qp_attributes attrs = {0}; CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep); if (ep && ep->com.qp) { printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", ep->hwtid, ep->com.qp->wq.sq.qid); attrs.next_state = C4IW_QP_STATE_TERMINATE; c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); } else printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", ep->hwtid); CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep); return 0; } int __init c4iw_cm_init(void) { t4_register_cpl_handler(CPL_RDMA_TERMINATE, terminate); t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, fw6_wr_rpl); t4_register_fw_msg_handler(FW6_TYPE_CQE, fw6_cqe_handler); t4_register_an_handler(c4iw_ev_handler); TAILQ_INIT(&req_list); spin_lock_init(&req_lock); INIT_LIST_HEAD(&err_cqe_list); spin_lock_init(&err_cqe_lock); INIT_WORK(&c4iw_task, process_req); c4iw_taskq = create_singlethread_workqueue("iw_cxgbe"); if (!c4iw_taskq) return -ENOMEM; return 0; } void __exit c4iw_cm_term(void) { WARN_ON(!TAILQ_EMPTY(&req_list)); WARN_ON(!list_empty(&err_cqe_list)); flush_workqueue(c4iw_taskq); destroy_workqueue(c4iw_taskq); t4_register_cpl_handler(CPL_RDMA_TERMINATE, NULL); t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, NULL); t4_register_fw_msg_handler(FW6_TYPE_CQE, NULL); t4_register_an_handler(NULL); } #endif Index: head/sys/dev/cxgbe/iw_cxgbe/qp.c =================================================================== --- head/sys/dev/cxgbe/iw_cxgbe/qp.c (revision 340062) +++ head/sys/dev/cxgbe/iw_cxgbe/qp.c (revision 340063) @@ -1,1951 +1,1967 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009-2013 Chelsio, Inc. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU * General Public License (GPL) Version 2, available from the file * COPYING in the main directory of this source tree, or the * OpenIB.org BSD license below: * * Redistribution and use in source and binary forms, with or * without modification, are permitted provided that the following * conditions are met: * * - Redistributions of source code must retain the above * copyright notice, this list of conditions and the following * disclaimer. * * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials * provided with the distribution. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #ifdef TCP_OFFLOAD #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include struct sge_iq; struct rss_header; struct cpl_set_tcb_rpl; #include #include "offload.h" #include "tom/t4_tom.h" #include "iw_cxgbe.h" #include "user.h" extern int use_dsgl; static int creds(struct toepcb *toep, struct inpcb *inp, size_t wrsize); static int max_fr_immd = T4_MAX_FR_IMMD;//SYSCTL parameter later... static int alloc_ird(struct c4iw_dev *dev, u32 ird) { int ret = 0; spin_lock_irq(&dev->lock); if (ird <= dev->avail_ird) dev->avail_ird -= ird; else ret = -ENOMEM; spin_unlock_irq(&dev->lock); if (ret) log(LOG_WARNING, "%s: device IRD resources exhausted\n", device_get_nameunit(dev->rdev.adap->dev)); return ret; } static void free_ird(struct c4iw_dev *dev, int ird) { spin_lock_irq(&dev->lock); dev->avail_ird += ird; spin_unlock_irq(&dev->lock); } static void set_state(struct c4iw_qp *qhp, enum c4iw_qp_state state) { unsigned long flag; spin_lock_irqsave(&qhp->lock, flag); qhp->attr.state = state; spin_unlock_irqrestore(&qhp->lock, flag); } static int destroy_qp(struct c4iw_rdev *rdev, struct t4_wq *wq, struct c4iw_dev_ucontext *uctx) { struct c4iw_dev *rhp = rdev_to_c4iw_dev(rdev); /* * uP clears EQ contexts when the connection exits rdma mode, * so no need to post a RESET WR for these EQs. */ dma_free_coherent(rhp->ibdev.dma_device, wq->rq.memsize, wq->rq.queue, dma_unmap_addr(&wq->rq, mapping)); dma_free_coherent(rhp->ibdev.dma_device, wq->sq.memsize, wq->sq.queue, dma_unmap_addr(&wq->sq, mapping)); c4iw_rqtpool_free(rdev, wq->rq.rqt_hwaddr, wq->rq.rqt_size); kfree(wq->rq.sw_rq); kfree(wq->sq.sw_sq); c4iw_put_qpid(rdev, wq->rq.qid, uctx); c4iw_put_qpid(rdev, wq->sq.qid, uctx); return 0; } static int create_qp(struct c4iw_rdev *rdev, struct t4_wq *wq, struct t4_cq *rcq, struct t4_cq *scq, struct c4iw_dev_ucontext *uctx) { struct adapter *sc = rdev->adap; struct c4iw_dev *rhp = rdev_to_c4iw_dev(rdev); int user = (uctx != &rdev->uctx); struct fw_ri_res_wr *res_wr; struct fw_ri_res *res; int wr_len; struct c4iw_wr_wait wr_wait; int ret = 0; int eqsize; struct wrqe *wr; u64 sq_bar2_qoffset = 0, rq_bar2_qoffset = 0; wq->sq.qid = c4iw_get_qpid(rdev, uctx); if (!wq->sq.qid) return -ENOMEM; wq->rq.qid = c4iw_get_qpid(rdev, uctx); if (!wq->rq.qid) { ret = -ENOMEM; goto free_sq_qid; } if (!user) { wq->sq.sw_sq = kzalloc(wq->sq.size * sizeof *wq->sq.sw_sq, GFP_KERNEL); if (!wq->sq.sw_sq) { ret = -ENOMEM; goto free_rq_qid; } wq->rq.sw_rq = kzalloc(wq->rq.size * sizeof *wq->rq.sw_rq, GFP_KERNEL); if (!wq->rq.sw_rq) { ret = -ENOMEM; goto free_sw_sq; } } /* * RQT must be a power of 2 and at least 16 deep. */ wq->rq.rqt_size = roundup_pow_of_two(max_t(u16, wq->rq.size, 16)); wq->rq.rqt_hwaddr = c4iw_rqtpool_alloc(rdev, wq->rq.rqt_size); if (!wq->rq.rqt_hwaddr) { ret = -ENOMEM; goto free_sw_rq; } /*QP memory, allocate DMAable memory for Send & Receive Queues */ wq->sq.queue = dma_alloc_coherent(rhp->ibdev.dma_device, wq->sq.memsize, &(wq->sq.dma_addr), GFP_KERNEL); if (!wq->sq.queue) { ret = -ENOMEM; goto free_hwaddr; } wq->sq.phys_addr = vtophys(wq->sq.queue); dma_unmap_addr_set(&wq->sq, mapping, wq->sq.dma_addr); memset(wq->sq.queue, 0, wq->sq.memsize); wq->rq.queue = dma_alloc_coherent(rhp->ibdev.dma_device, wq->rq.memsize, &(wq->rq.dma_addr), GFP_KERNEL); if (!wq->rq.queue) { ret = -ENOMEM; goto free_sq_dma; } wq->rq.phys_addr = vtophys(wq->rq.queue); dma_unmap_addr_set(&wq->rq, mapping, wq->rq.dma_addr); memset(wq->rq.queue, 0, wq->rq.memsize); CTR5(KTR_IW_CXGBE, "%s QP sq base va 0x%p pa 0x%llx rq base va 0x%p pa 0x%llx", __func__, wq->sq.queue, (unsigned long long)wq->sq.phys_addr, wq->rq.queue, (unsigned long long)wq->rq.phys_addr); /* Doorbell/WC regions, determine the BAR2 queue offset and qid. */ t4_bar2_sge_qregs(rdev->adap, wq->sq.qid, T4_BAR2_QTYPE_EGRESS, user, &sq_bar2_qoffset, &wq->sq.bar2_qid); t4_bar2_sge_qregs(rdev->adap, wq->rq.qid, T4_BAR2_QTYPE_EGRESS, user, &rq_bar2_qoffset, &wq->rq.bar2_qid); if (user) { /* Compute BAR2 DB/WC physical address(page-aligned) for * Userspace mapping. */ wq->sq.bar2_pa = (rdev->bar2_pa + sq_bar2_qoffset) & PAGE_MASK; wq->rq.bar2_pa = (rdev->bar2_pa + rq_bar2_qoffset) & PAGE_MASK; CTR3(KTR_IW_CXGBE, "%s BAR2 DB/WC sq base pa 0x%llx rq base pa 0x%llx", __func__, (unsigned long long)wq->sq.bar2_pa, (unsigned long long)wq->rq.bar2_pa); } else { /* Compute BAR2 DB/WC virtual address to access in kernel. */ wq->sq.bar2_va = (void __iomem *)((u64)rdev->bar2_kva + sq_bar2_qoffset); wq->rq.bar2_va = (void __iomem *)((u64)rdev->bar2_kva + rq_bar2_qoffset); CTR3(KTR_IW_CXGBE, "%s BAR2 DB/WC sq base va %p rq base va %p", __func__, (unsigned long long)wq->sq.bar2_va, (unsigned long long)wq->rq.bar2_va); } wq->rdev = rdev; wq->rq.msn = 1; /* build fw_ri_res_wr */ wr_len = sizeof *res_wr + 2 * sizeof *res; wr = alloc_wrqe(wr_len, &sc->sge.ctrlq[0]); if (wr == NULL) { ret = -ENOMEM; goto free_rq_dma; } res_wr = wrtod(wr); memset(res_wr, 0, wr_len); res_wr->op_nres = cpu_to_be32( V_FW_WR_OP(FW_RI_RES_WR) | V_FW_RI_RES_WR_NRES(2) | F_FW_WR_COMPL); res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16)); res_wr->cookie = (unsigned long) &wr_wait; res = res_wr->res; res->u.sqrq.restype = FW_RI_RES_TYPE_SQ; res->u.sqrq.op = FW_RI_RES_OP_WRITE; /* eqsize is the number of 64B entries plus the status page size. */ eqsize = wq->sq.size * T4_SQ_NUM_SLOTS + rdev->hw_queue.t4_eq_status_entries; res->u.sqrq.fetchszm_to_iqid = cpu_to_be32( V_FW_RI_RES_WR_HOSTFCMODE(0) | /* no host cidx updates */ V_FW_RI_RES_WR_CPRIO(0) | /* don't keep in chip cache */ V_FW_RI_RES_WR_PCIECHN(0) | /* set by uP at ri_init time */ V_FW_RI_RES_WR_IQID(scq->cqid)); res->u.sqrq.dcaen_to_eqsize = cpu_to_be32( V_FW_RI_RES_WR_DCAEN(0) | V_FW_RI_RES_WR_DCACPU(0) | V_FW_RI_RES_WR_FBMIN(2) | V_FW_RI_RES_WR_FBMAX(3) | V_FW_RI_RES_WR_CIDXFTHRESHO(0) | V_FW_RI_RES_WR_CIDXFTHRESH(0) | V_FW_RI_RES_WR_EQSIZE(eqsize)); res->u.sqrq.eqid = cpu_to_be32(wq->sq.qid); res->u.sqrq.eqaddr = cpu_to_be64(wq->sq.dma_addr); res++; res->u.sqrq.restype = FW_RI_RES_TYPE_RQ; res->u.sqrq.op = FW_RI_RES_OP_WRITE; /* eqsize is the number of 64B entries plus the status page size. */ eqsize = wq->rq.size * T4_RQ_NUM_SLOTS + rdev->hw_queue.t4_eq_status_entries; res->u.sqrq.fetchszm_to_iqid = cpu_to_be32( V_FW_RI_RES_WR_HOSTFCMODE(0) | /* no host cidx updates */ V_FW_RI_RES_WR_CPRIO(0) | /* don't keep in chip cache */ V_FW_RI_RES_WR_PCIECHN(0) | /* set by uP at ri_init time */ V_FW_RI_RES_WR_IQID(rcq->cqid)); res->u.sqrq.dcaen_to_eqsize = cpu_to_be32( V_FW_RI_RES_WR_DCAEN(0) | V_FW_RI_RES_WR_DCACPU(0) | V_FW_RI_RES_WR_FBMIN(2) | V_FW_RI_RES_WR_FBMAX(3) | V_FW_RI_RES_WR_CIDXFTHRESHO(0) | V_FW_RI_RES_WR_CIDXFTHRESH(0) | V_FW_RI_RES_WR_EQSIZE(eqsize)); res->u.sqrq.eqid = cpu_to_be32(wq->rq.qid); res->u.sqrq.eqaddr = cpu_to_be64(wq->rq.dma_addr); c4iw_init_wr_wait(&wr_wait); t4_wrq_tx(sc, wr); ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, wq->sq.qid, NULL, __func__); if (ret) goto free_rq_dma; CTR5(KTR_IW_CXGBE, "%s sqid 0x%x rqid 0x%x kdb 0x%p squdb 0x%llx rqudb 0x%llx", __func__, wq->sq.qid, wq->rq.qid, (unsigned long long)wq->sq.bar2_va, (unsigned long long)wq->rq.bar2_va); return 0; free_rq_dma: dma_free_coherent(rhp->ibdev.dma_device, wq->rq.memsize, wq->rq.queue, dma_unmap_addr(&wq->rq, mapping)); free_sq_dma: dma_free_coherent(rhp->ibdev.dma_device, wq->sq.memsize, wq->sq.queue, dma_unmap_addr(&wq->sq, mapping)); free_hwaddr: c4iw_rqtpool_free(rdev, wq->rq.rqt_hwaddr, wq->rq.rqt_size); free_sw_rq: kfree(wq->rq.sw_rq); free_sw_sq: kfree(wq->sq.sw_sq); free_rq_qid: c4iw_put_qpid(rdev, wq->rq.qid, uctx); free_sq_qid: c4iw_put_qpid(rdev, wq->sq.qid, uctx); return ret; } static int build_immd(struct t4_sq *sq, struct fw_ri_immd *immdp, struct ib_send_wr *wr, int max, u32 *plenp) { u8 *dstp, *srcp; u32 plen = 0; int i; int rem, len; dstp = (u8 *)immdp->data; for (i = 0; i < wr->num_sge; i++) { if ((plen + wr->sg_list[i].length) > max) return -EMSGSIZE; srcp = (u8 *)(unsigned long)wr->sg_list[i].addr; plen += wr->sg_list[i].length; rem = wr->sg_list[i].length; while (rem) { if (dstp == (u8 *)&sq->queue[sq->size]) dstp = (u8 *)sq->queue; if (rem <= (u8 *)&sq->queue[sq->size] - dstp) len = rem; else len = (u8 *)&sq->queue[sq->size] - dstp; memcpy(dstp, srcp, len); dstp += len; srcp += len; rem -= len; } } len = roundup(plen + sizeof *immdp, 16) - (plen + sizeof *immdp); if (len) memset(dstp, 0, len); immdp->op = FW_RI_DATA_IMMD; immdp->r1 = 0; immdp->r2 = 0; immdp->immdlen = cpu_to_be32(plen); *plenp = plen; return 0; } static int build_isgl(__be64 *queue_start, __be64 *queue_end, struct fw_ri_isgl *isglp, struct ib_sge *sg_list, int num_sge, u32 *plenp) { int i; u32 plen = 0; __be64 *flitp = (__be64 *)isglp->sge; for (i = 0; i < num_sge; i++) { if ((plen + sg_list[i].length) < plen) return -EMSGSIZE; plen += sg_list[i].length; *flitp = cpu_to_be64(((u64)sg_list[i].lkey << 32) | sg_list[i].length); if (++flitp == queue_end) flitp = queue_start; *flitp = cpu_to_be64(sg_list[i].addr); if (++flitp == queue_end) flitp = queue_start; } *flitp = (__force __be64)0; isglp->op = FW_RI_DATA_ISGL; isglp->r1 = 0; isglp->nsge = cpu_to_be16(num_sge); isglp->r2 = 0; if (plenp) *plenp = plen; return 0; } static int build_rdma_send(struct t4_sq *sq, union t4_wr *wqe, struct ib_send_wr *wr, u8 *len16) { u32 plen; int size; int ret; if (wr->num_sge > T4_MAX_SEND_SGE) return -EINVAL; switch (wr->opcode) { case IB_WR_SEND: if (wr->send_flags & IB_SEND_SOLICITED) wqe->send.sendop_pkd = cpu_to_be32( V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND_WITH_SE)); else wqe->send.sendop_pkd = cpu_to_be32( V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND)); wqe->send.stag_inv = 0; break; case IB_WR_SEND_WITH_INV: if (wr->send_flags & IB_SEND_SOLICITED) wqe->send.sendop_pkd = cpu_to_be32( V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND_WITH_SE_INV)); else wqe->send.sendop_pkd = cpu_to_be32( V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND_WITH_INV)); wqe->send.stag_inv = cpu_to_be32(wr->ex.invalidate_rkey); break; default: return -EINVAL; } wqe->send.r3 = 0; wqe->send.r4 = 0; plen = 0; if (wr->num_sge) { if (wr->send_flags & IB_SEND_INLINE) { ret = build_immd(sq, wqe->send.u.immd_src, wr, T4_MAX_SEND_INLINE, &plen); if (ret) return ret; size = sizeof wqe->send + sizeof(struct fw_ri_immd) + plen; } else { ret = build_isgl((__be64 *)sq->queue, (__be64 *)&sq->queue[sq->size], wqe->send.u.isgl_src, wr->sg_list, wr->num_sge, &plen); if (ret) return ret; size = sizeof wqe->send + sizeof(struct fw_ri_isgl) + wr->num_sge * sizeof(struct fw_ri_sge); } } else { wqe->send.u.immd_src[0].op = FW_RI_DATA_IMMD; wqe->send.u.immd_src[0].r1 = 0; wqe->send.u.immd_src[0].r2 = 0; wqe->send.u.immd_src[0].immdlen = 0; size = sizeof wqe->send + sizeof(struct fw_ri_immd); plen = 0; } *len16 = DIV_ROUND_UP(size, 16); wqe->send.plen = cpu_to_be32(plen); return 0; } static int build_rdma_write(struct t4_sq *sq, union t4_wr *wqe, struct ib_send_wr *wr, u8 *len16) { u32 plen; int size; int ret; if (wr->num_sge > T4_MAX_SEND_SGE) return -EINVAL; wqe->write.immd_data = 0; wqe->write.stag_sink = cpu_to_be32(rdma_wr(wr)->rkey); wqe->write.to_sink = cpu_to_be64(rdma_wr(wr)->remote_addr); if (wr->num_sge) { if (wr->send_flags & IB_SEND_INLINE) { ret = build_immd(sq, wqe->write.u.immd_src, wr, T4_MAX_WRITE_INLINE, &plen); if (ret) return ret; size = sizeof wqe->write + sizeof(struct fw_ri_immd) + plen; } else { ret = build_isgl((__be64 *)sq->queue, (__be64 *)&sq->queue[sq->size], wqe->write.u.isgl_src, wr->sg_list, wr->num_sge, &plen); if (ret) return ret; size = sizeof wqe->write + sizeof(struct fw_ri_isgl) + wr->num_sge * sizeof(struct fw_ri_sge); } } else { wqe->write.u.immd_src[0].op = FW_RI_DATA_IMMD; wqe->write.u.immd_src[0].r1 = 0; wqe->write.u.immd_src[0].r2 = 0; wqe->write.u.immd_src[0].immdlen = 0; size = sizeof wqe->write + sizeof(struct fw_ri_immd); plen = 0; } *len16 = DIV_ROUND_UP(size, 16); wqe->write.plen = cpu_to_be32(plen); return 0; } static int build_rdma_read(union t4_wr *wqe, struct ib_send_wr *wr, u8 *len16) { if (wr->num_sge > 1) return -EINVAL; if (wr->num_sge && wr->sg_list[0].length) { wqe->read.stag_src = cpu_to_be32(rdma_wr(wr)->rkey); wqe->read.to_src_hi = cpu_to_be32((u32)(rdma_wr(wr)->remote_addr >> 32)); wqe->read.to_src_lo = cpu_to_be32((u32)rdma_wr(wr)->remote_addr); wqe->read.stag_sink = cpu_to_be32(wr->sg_list[0].lkey); wqe->read.plen = cpu_to_be32(wr->sg_list[0].length); wqe->read.to_sink_hi = cpu_to_be32((u32)(wr->sg_list[0].addr >> 32)); wqe->read.to_sink_lo = cpu_to_be32((u32)(wr->sg_list[0].addr)); } else { wqe->read.stag_src = cpu_to_be32(2); wqe->read.to_src_hi = 0; wqe->read.to_src_lo = 0; wqe->read.stag_sink = cpu_to_be32(2); wqe->read.plen = 0; wqe->read.to_sink_hi = 0; wqe->read.to_sink_lo = 0; } wqe->read.r2 = 0; wqe->read.r5 = 0; *len16 = DIV_ROUND_UP(sizeof wqe->read, 16); return 0; } static int build_rdma_recv(struct c4iw_qp *qhp, union t4_recv_wr *wqe, struct ib_recv_wr *wr, u8 *len16) { int ret; ret = build_isgl((__be64 *)qhp->wq.rq.queue, (__be64 *)&qhp->wq.rq.queue[qhp->wq.rq.size], &wqe->recv.isgl, wr->sg_list, wr->num_sge, NULL); if (ret) return ret; *len16 = DIV_ROUND_UP(sizeof wqe->recv + wr->num_sge * sizeof(struct fw_ri_sge), 16); return 0; } static int build_inv_stag(union t4_wr *wqe, struct ib_send_wr *wr, u8 *len16) { wqe->inv.stag_inv = cpu_to_be32(wr->ex.invalidate_rkey); wqe->inv.r2 = 0; *len16 = DIV_ROUND_UP(sizeof wqe->inv, 16); return 0; } static void free_qp_work(struct work_struct *work) { struct c4iw_ucontext *ucontext; struct c4iw_qp *qhp; struct c4iw_dev *rhp; qhp = container_of(work, struct c4iw_qp, free_work); ucontext = qhp->ucontext; rhp = qhp->rhp; CTR3(KTR_IW_CXGBE, "%s qhp %p ucontext %p\n", __func__, qhp, ucontext); destroy_qp(&rhp->rdev, &qhp->wq, ucontext ? &ucontext->uctx : &rhp->rdev.uctx); if (ucontext) c4iw_put_ucontext(ucontext); kfree(qhp); } static void queue_qp_free(struct kref *kref) { struct c4iw_qp *qhp; qhp = container_of(kref, struct c4iw_qp, kref); CTR2(KTR_IW_CXGBE, "%s qhp %p", __func__, qhp); queue_work(qhp->rhp->rdev.free_workq, &qhp->free_work); } void c4iw_qp_add_ref(struct ib_qp *qp) { CTR2(KTR_IW_CXGBE, "%s ib_qp %p", __func__, qp); kref_get(&to_c4iw_qp(qp)->kref); } void c4iw_qp_rem_ref(struct ib_qp *qp) { CTR2(KTR_IW_CXGBE, "%s ib_qp %p", __func__, qp); kref_put(&to_c4iw_qp(qp)->kref, queue_qp_free); } static void complete_sq_drain_wr(struct c4iw_qp *qhp, struct ib_send_wr *wr) { struct t4_cqe cqe = {}; struct c4iw_cq *schp; unsigned long flag; struct t4_cq *cq; schp = to_c4iw_cq(qhp->ibqp.send_cq); cq = &schp->cq; PDBG("%s drain sq id %u\n", __func__, qhp->wq.sq.qid); cqe.u.drain_cookie = wr->wr_id; cqe.header = cpu_to_be32(V_CQE_STATUS(T4_ERR_SWFLUSH) | V_CQE_OPCODE(C4IW_DRAIN_OPCODE) | V_CQE_TYPE(1) | V_CQE_SWCQE(1) | V_CQE_QPID(qhp->wq.sq.qid)); spin_lock_irqsave(&schp->lock, flag); cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen)); cq->sw_queue[cq->sw_pidx] = cqe; t4_swcq_produce(cq); spin_unlock_irqrestore(&schp->lock, flag); spin_lock_irqsave(&schp->comp_handler_lock, flag); (*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context); spin_unlock_irqrestore(&schp->comp_handler_lock, flag); } static void complete_rq_drain_wr(struct c4iw_qp *qhp, struct ib_recv_wr *wr) { struct t4_cqe cqe = {}; struct c4iw_cq *rchp; unsigned long flag; struct t4_cq *cq; rchp = to_c4iw_cq(qhp->ibqp.recv_cq); cq = &rchp->cq; PDBG("%s drain rq id %u\n", __func__, qhp->wq.sq.qid); cqe.u.drain_cookie = wr->wr_id; cqe.header = cpu_to_be32(V_CQE_STATUS(T4_ERR_SWFLUSH) | V_CQE_OPCODE(C4IW_DRAIN_OPCODE) | V_CQE_TYPE(0) | V_CQE_SWCQE(1) | V_CQE_QPID(qhp->wq.sq.qid)); spin_lock_irqsave(&rchp->lock, flag); cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen)); cq->sw_queue[cq->sw_pidx] = cqe; t4_swcq_produce(cq); spin_unlock_irqrestore(&rchp->lock, flag); spin_lock_irqsave(&rchp->comp_handler_lock, flag); (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context); spin_unlock_irqrestore(&rchp->comp_handler_lock, flag); } static void build_tpte_memreg(struct fw_ri_fr_nsmr_tpte_wr *fr, struct ib_reg_wr *wr, struct c4iw_mr *mhp, u8 *len16) { __be64 *p = (__be64 *)fr->pbl; fr->r2 = cpu_to_be32(0); fr->stag = cpu_to_be32(mhp->ibmr.rkey); fr->tpte.valid_to_pdid = cpu_to_be32(F_FW_RI_TPTE_VALID | V_FW_RI_TPTE_STAGKEY((mhp->ibmr.rkey & M_FW_RI_TPTE_STAGKEY)) | V_FW_RI_TPTE_STAGSTATE(1) | V_FW_RI_TPTE_STAGTYPE(FW_RI_STAG_NSMR) | V_FW_RI_TPTE_PDID(mhp->attr.pdid)); fr->tpte.locread_to_qpid = cpu_to_be32( V_FW_RI_TPTE_PERM(c4iw_ib_to_tpt_access(wr->access)) | V_FW_RI_TPTE_ADDRTYPE(FW_RI_VA_BASED_TO) | V_FW_RI_TPTE_PS(ilog2(wr->mr->page_size) - 12)); fr->tpte.nosnoop_pbladdr = cpu_to_be32(V_FW_RI_TPTE_PBLADDR( PBL_OFF(&mhp->rhp->rdev, mhp->attr.pbl_addr)>>3)); fr->tpte.dca_mwbcnt_pstag = cpu_to_be32(0); fr->tpte.len_hi = cpu_to_be32(0); fr->tpte.len_lo = cpu_to_be32(mhp->ibmr.length); fr->tpte.va_hi = cpu_to_be32(mhp->ibmr.iova >> 32); fr->tpte.va_lo_fbo = cpu_to_be32(mhp->ibmr.iova & 0xffffffff); p[0] = cpu_to_be64((u64)mhp->mpl[0]); p[1] = cpu_to_be64((u64)mhp->mpl[1]); *len16 = DIV_ROUND_UP(sizeof(*fr), 16); } static int build_memreg(struct t4_sq *sq, union t4_wr *wqe, struct ib_reg_wr *wr, struct c4iw_mr *mhp, u8 *len16, bool dsgl_supported) { struct fw_ri_immd *imdp; __be64 *p; int i; int pbllen = roundup(mhp->mpl_len * sizeof(u64), 32); int rem; if (mhp->mpl_len > t4_max_fr_depth(use_dsgl && dsgl_supported)) return -EINVAL; wqe->fr.qpbinde_to_dcacpu = 0; wqe->fr.pgsz_shift = ilog2(wr->mr->page_size) - 12; wqe->fr.addr_type = FW_RI_VA_BASED_TO; wqe->fr.mem_perms = c4iw_ib_to_tpt_access(wr->access); wqe->fr.len_hi = 0; wqe->fr.len_lo = cpu_to_be32(mhp->ibmr.length); wqe->fr.stag = cpu_to_be32(wr->key); wqe->fr.va_hi = cpu_to_be32(mhp->ibmr.iova >> 32); wqe->fr.va_lo_fbo = cpu_to_be32(mhp->ibmr.iova & 0xffffffff); if (dsgl_supported && use_dsgl && (pbllen > max_fr_immd)) { struct fw_ri_dsgl *sglp; for (i = 0; i < mhp->mpl_len; i++) mhp->mpl[i] = (__force u64)cpu_to_be64((u64)mhp->mpl[i]); sglp = (struct fw_ri_dsgl *)(&wqe->fr + 1); sglp->op = FW_RI_DATA_DSGL; sglp->r1 = 0; sglp->nsge = cpu_to_be16(1); sglp->addr0 = cpu_to_be64(mhp->mpl_addr); sglp->len0 = cpu_to_be32(pbllen); *len16 = DIV_ROUND_UP(sizeof(wqe->fr) + sizeof(*sglp), 16); } else { imdp = (struct fw_ri_immd *)(&wqe->fr + 1); imdp->op = FW_RI_DATA_IMMD; imdp->r1 = 0; imdp->r2 = 0; imdp->immdlen = cpu_to_be32(pbllen); p = (__be64 *)(imdp + 1); rem = pbllen; for (i = 0; i < mhp->mpl_len; i++) { *p = cpu_to_be64((u64)mhp->mpl[i]); rem -= sizeof(*p); if (++p == (__be64 *)&sq->queue[sq->size]) p = (__be64 *)sq->queue; } BUG_ON(rem < 0); while (rem) { *p = 0; rem -= sizeof(*p); if (++p == (__be64 *)&sq->queue[sq->size]) p = (__be64 *)sq->queue; } *len16 = DIV_ROUND_UP(sizeof(wqe->fr) + sizeof(*imdp) + pbllen, 16); } return 0; } int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, struct ib_send_wr **bad_wr) { int err = 0; u8 len16 = 0; enum fw_wr_opcodes fw_opcode = 0; enum fw_ri_wr_flags fw_flags; struct c4iw_qp *qhp; union t4_wr *wqe = NULL; u32 num_wrs; struct t4_swsqe *swsqe; unsigned long flag; u16 idx = 0; struct c4iw_rdev *rdev; qhp = to_c4iw_qp(ibqp); rdev = &qhp->rhp->rdev; spin_lock_irqsave(&qhp->lock, flag); if (t4_wq_in_error(&qhp->wq)) { spin_unlock_irqrestore(&qhp->lock, flag); complete_sq_drain_wr(qhp, wr); return err; } num_wrs = t4_sq_avail(&qhp->wq); if (num_wrs == 0) { spin_unlock_irqrestore(&qhp->lock, flag); *bad_wr = wr; return -ENOMEM; } while (wr) { if (num_wrs == 0) { err = -ENOMEM; *bad_wr = wr; break; } wqe = (union t4_wr *)((u8 *)qhp->wq.sq.queue + qhp->wq.sq.wq_pidx * T4_EQ_ENTRY_SIZE); fw_flags = 0; if (wr->send_flags & IB_SEND_SOLICITED) fw_flags |= FW_RI_SOLICITED_EVENT_FLAG; if (wr->send_flags & IB_SEND_SIGNALED || qhp->sq_sig_all) fw_flags |= FW_RI_COMPLETION_FLAG; swsqe = &qhp->wq.sq.sw_sq[qhp->wq.sq.pidx]; switch (wr->opcode) { case IB_WR_SEND_WITH_INV: case IB_WR_SEND: if (wr->send_flags & IB_SEND_FENCE) fw_flags |= FW_RI_READ_FENCE_FLAG; fw_opcode = FW_RI_SEND_WR; if (wr->opcode == IB_WR_SEND) swsqe->opcode = FW_RI_SEND; else swsqe->opcode = FW_RI_SEND_WITH_INV; err = build_rdma_send(&qhp->wq.sq, wqe, wr, &len16); break; case IB_WR_RDMA_WRITE: fw_opcode = FW_RI_RDMA_WRITE_WR; swsqe->opcode = FW_RI_RDMA_WRITE; err = build_rdma_write(&qhp->wq.sq, wqe, wr, &len16); break; case IB_WR_RDMA_READ: case IB_WR_RDMA_READ_WITH_INV: fw_opcode = FW_RI_RDMA_READ_WR; swsqe->opcode = FW_RI_READ_REQ; if (wr->opcode == IB_WR_RDMA_READ_WITH_INV) { c4iw_invalidate_mr(qhp->rhp, wr->sg_list[0].lkey); fw_flags = FW_RI_RDMA_READ_INVALIDATE; } else { fw_flags = 0; } err = build_rdma_read(wqe, wr, &len16); if (err) break; swsqe->read_len = wr->sg_list[0].length; if (!qhp->wq.sq.oldest_read) qhp->wq.sq.oldest_read = swsqe; break; case IB_WR_REG_MR: { struct c4iw_mr *mhp = to_c4iw_mr(reg_wr(wr)->mr); swsqe->opcode = FW_RI_FAST_REGISTER; if (rdev->adap->params.fr_nsmr_tpte_wr_support && !mhp->attr.state && mhp->mpl_len <= 2) { fw_opcode = FW_RI_FR_NSMR_TPTE_WR; build_tpte_memreg(&wqe->fr_tpte, reg_wr(wr), mhp, &len16); } else { fw_opcode = FW_RI_FR_NSMR_WR; err = build_memreg(&qhp->wq.sq, wqe, reg_wr(wr), mhp, &len16, rdev->adap->params.ulptx_memwrite_dsgl); if (err) break; } mhp->attr.state = 1; break; } case IB_WR_LOCAL_INV: if (wr->send_flags & IB_SEND_FENCE) fw_flags |= FW_RI_LOCAL_FENCE_FLAG; fw_opcode = FW_RI_INV_LSTAG_WR; swsqe->opcode = FW_RI_LOCAL_INV; err = build_inv_stag(wqe, wr, &len16); c4iw_invalidate_mr(qhp->rhp, wr->ex.invalidate_rkey); break; default: CTR2(KTR_IW_CXGBE, "%s post of type =%d TBD!", __func__, wr->opcode); err = -EINVAL; } if (err) { *bad_wr = wr; break; } swsqe->idx = qhp->wq.sq.pidx; swsqe->complete = 0; swsqe->signaled = (wr->send_flags & IB_SEND_SIGNALED) || qhp->sq_sig_all; swsqe->flushed = 0; swsqe->wr_id = wr->wr_id; init_wr_hdr(wqe, qhp->wq.sq.pidx, fw_opcode, fw_flags, len16); CTR5(KTR_IW_CXGBE, "%s cookie 0x%llx pidx 0x%x opcode 0x%x read_len %u", __func__, (unsigned long long)wr->wr_id, qhp->wq.sq.pidx, swsqe->opcode, swsqe->read_len); wr = wr->next; num_wrs--; t4_sq_produce(&qhp->wq, len16); idx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE); } t4_ring_sq_db(&qhp->wq, idx, wqe, rdev->adap->iwt.wc_en); spin_unlock_irqrestore(&qhp->lock, flag); return err; } int c4iw_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr, struct ib_recv_wr **bad_wr) { int err = 0; struct c4iw_qp *qhp; union t4_recv_wr *wqe = NULL; u32 num_wrs; u8 len16 = 0; unsigned long flag; u16 idx = 0; qhp = to_c4iw_qp(ibqp); spin_lock_irqsave(&qhp->lock, flag); if (t4_wq_in_error(&qhp->wq)) { spin_unlock_irqrestore(&qhp->lock, flag); complete_rq_drain_wr(qhp, wr); return err; } num_wrs = t4_rq_avail(&qhp->wq); if (num_wrs == 0) { spin_unlock_irqrestore(&qhp->lock, flag); *bad_wr = wr; return -ENOMEM; } while (wr) { if (wr->num_sge > T4_MAX_RECV_SGE) { err = -EINVAL; *bad_wr = wr; break; } wqe = (union t4_recv_wr *)((u8 *)qhp->wq.rq.queue + qhp->wq.rq.wq_pidx * T4_EQ_ENTRY_SIZE); if (num_wrs) err = build_rdma_recv(qhp, wqe, wr, &len16); else err = -ENOMEM; if (err) { *bad_wr = wr; break; } qhp->wq.rq.sw_rq[qhp->wq.rq.pidx].wr_id = wr->wr_id; wqe->recv.opcode = FW_RI_RECV_WR; wqe->recv.r1 = 0; wqe->recv.wrid = qhp->wq.rq.pidx; wqe->recv.r2[0] = 0; wqe->recv.r2[1] = 0; wqe->recv.r2[2] = 0; wqe->recv.len16 = len16; CTR3(KTR_IW_CXGBE, "%s cookie 0x%llx pidx %u", __func__, (unsigned long long) wr->wr_id, qhp->wq.rq.pidx); t4_rq_produce(&qhp->wq, len16); idx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE); wr = wr->next; num_wrs--; } t4_ring_rq_db(&qhp->wq, idx, wqe, qhp->rhp->rdev.adap->iwt.wc_en); spin_unlock_irqrestore(&qhp->lock, flag); return err; } static inline void build_term_codes(struct t4_cqe *err_cqe, u8 *layer_type, u8 *ecode) { int status; int tagged; int opcode; int rqtype; int send_inv; if (!err_cqe) { *layer_type = LAYER_RDMAP|DDP_LOCAL_CATA; *ecode = 0; return; } status = CQE_STATUS(err_cqe); opcode = CQE_OPCODE(err_cqe); rqtype = RQ_TYPE(err_cqe); send_inv = (opcode == FW_RI_SEND_WITH_INV) || (opcode == FW_RI_SEND_WITH_SE_INV); tagged = (opcode == FW_RI_RDMA_WRITE) || (rqtype && (opcode == FW_RI_READ_RESP)); switch (status) { case T4_ERR_STAG: if (send_inv) { *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP; *ecode = RDMAP_CANT_INV_STAG; } else { *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT; *ecode = RDMAP_INV_STAG; } break; case T4_ERR_PDID: *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT; if ((opcode == FW_RI_SEND_WITH_INV) || (opcode == FW_RI_SEND_WITH_SE_INV)) *ecode = RDMAP_CANT_INV_STAG; else *ecode = RDMAP_STAG_NOT_ASSOC; break; case T4_ERR_QPID: *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT; *ecode = RDMAP_STAG_NOT_ASSOC; break; case T4_ERR_ACCESS: *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT; *ecode = RDMAP_ACC_VIOL; break; case T4_ERR_WRAP: *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT; *ecode = RDMAP_TO_WRAP; break; case T4_ERR_BOUND: if (tagged) { *layer_type = LAYER_DDP|DDP_TAGGED_ERR; *ecode = DDPT_BASE_BOUNDS; } else { *layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT; *ecode = RDMAP_BASE_BOUNDS; } break; case T4_ERR_INVALIDATE_SHARED_MR: case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND: *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP; *ecode = RDMAP_CANT_INV_STAG; break; case T4_ERR_ECC: case T4_ERR_ECC_PSTAG: case T4_ERR_INTERNAL_ERR: *layer_type = LAYER_RDMAP|RDMAP_LOCAL_CATA; *ecode = 0; break; case T4_ERR_OUT_OF_RQE: *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR; *ecode = DDPU_INV_MSN_NOBUF; break; case T4_ERR_PBL_ADDR_BOUND: *layer_type = LAYER_DDP|DDP_TAGGED_ERR; *ecode = DDPT_BASE_BOUNDS; break; case T4_ERR_CRC: *layer_type = LAYER_MPA|DDP_LLP; *ecode = MPA_CRC_ERR; break; case T4_ERR_MARKER: *layer_type = LAYER_MPA|DDP_LLP; *ecode = MPA_MARKER_ERR; break; case T4_ERR_PDU_LEN_ERR: *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR; *ecode = DDPU_MSG_TOOBIG; break; case T4_ERR_DDP_VERSION: if (tagged) { *layer_type = LAYER_DDP|DDP_TAGGED_ERR; *ecode = DDPT_INV_VERS; } else { *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR; *ecode = DDPU_INV_VERS; } break; case T4_ERR_RDMA_VERSION: *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP; *ecode = RDMAP_INV_VERS; break; case T4_ERR_OPCODE: *layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP; *ecode = RDMAP_INV_OPCODE; break; case T4_ERR_DDP_QUEUE_NUM: *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR; *ecode = DDPU_INV_QN; break; case T4_ERR_MSN: case T4_ERR_MSN_GAP: case T4_ERR_MSN_RANGE: case T4_ERR_IRD_OVERFLOW: *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR; *ecode = DDPU_INV_MSN_RANGE; break; case T4_ERR_TBIT: *layer_type = LAYER_DDP|DDP_LOCAL_CATA; *ecode = 0; break; case T4_ERR_MO: *layer_type = LAYER_DDP|DDP_UNTAGGED_ERR; *ecode = DDPU_INV_MO; break; default: *layer_type = LAYER_RDMAP|DDP_LOCAL_CATA; *ecode = 0; break; } } static void post_terminate(struct c4iw_qp *qhp, struct t4_cqe *err_cqe, gfp_t gfp) { int ret; struct fw_ri_wr *wqe; struct terminate_message *term; struct wrqe *wr; struct socket *so = qhp->ep->com.so; struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = intotcpcb(inp); struct toepcb *toep = tp->t_toe; CTR4(KTR_IW_CXGBE, "%s qhp %p qid 0x%x tid %u", __func__, qhp, qhp->wq.sq.qid, qhp->ep->hwtid); wr = alloc_wrqe(sizeof(*wqe), toep->ofld_txq); if (wr == NULL) return; wqe = wrtod(wr); memset(wqe, 0, sizeof *wqe); wqe->op_compl = cpu_to_be32(V_FW_WR_OP(FW_RI_WR)); wqe->flowid_len16 = cpu_to_be32( V_FW_WR_FLOWID(qhp->ep->hwtid) | V_FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe, 16))); wqe->u.terminate.type = FW_RI_TYPE_TERMINATE; wqe->u.terminate.immdlen = cpu_to_be32(sizeof *term); term = (struct terminate_message *)wqe->u.terminate.termmsg; if (qhp->attr.layer_etype == (LAYER_MPA|DDP_LLP)) { term->layer_etype = qhp->attr.layer_etype; term->ecode = qhp->attr.ecode; } else build_term_codes(err_cqe, &term->layer_etype, &term->ecode); ret = creds(toep, inp, sizeof(*wqe)); if (ret) { free_wrqe(wr); return; } t4_wrq_tx(qhp->rhp->rdev.adap, wr); } /* Assumes qhp lock is held. */ static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp, struct c4iw_cq *schp) { int count; int rq_flushed, sq_flushed; unsigned long flag; CTR4(KTR_IW_CXGBE, "%s qhp %p rchp %p schp %p", __func__, qhp, rchp, schp); /* locking hierarchy: cq lock first, then qp lock. */ spin_lock_irqsave(&rchp->lock, flag); spin_lock(&qhp->lock); if (qhp->wq.flushed) { spin_unlock(&qhp->lock); spin_unlock_irqrestore(&rchp->lock, flag); return; } qhp->wq.flushed = 1; c4iw_flush_hw_cq(rchp); c4iw_count_rcqes(&rchp->cq, &qhp->wq, &count); rq_flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count); spin_unlock(&qhp->lock); spin_unlock_irqrestore(&rchp->lock, flag); /* locking hierarchy: cq lock first, then qp lock. */ spin_lock_irqsave(&schp->lock, flag); spin_lock(&qhp->lock); if (schp != rchp) c4iw_flush_hw_cq(schp); sq_flushed = c4iw_flush_sq(qhp); spin_unlock(&qhp->lock); spin_unlock_irqrestore(&schp->lock, flag); if (schp == rchp) { if (t4_clear_cq_armed(&rchp->cq) && (rq_flushed || sq_flushed)) { spin_lock_irqsave(&rchp->comp_handler_lock, flag); (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context); spin_unlock_irqrestore(&rchp->comp_handler_lock, flag); } } else { if (t4_clear_cq_armed(&rchp->cq) && rq_flushed) { spin_lock_irqsave(&rchp->comp_handler_lock, flag); (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context); spin_unlock_irqrestore(&rchp->comp_handler_lock, flag); } if (t4_clear_cq_armed(&schp->cq) && sq_flushed) { spin_lock_irqsave(&schp->comp_handler_lock, flag); (*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context); spin_unlock_irqrestore(&schp->comp_handler_lock, flag); } } } static void flush_qp(struct c4iw_qp *qhp) { struct c4iw_cq *rchp, *schp; unsigned long flag; rchp = to_c4iw_cq(qhp->ibqp.recv_cq); schp = to_c4iw_cq(qhp->ibqp.send_cq); t4_set_wq_in_error(&qhp->wq); if (qhp->ibqp.uobject) { t4_set_cq_in_error(&rchp->cq); spin_lock_irqsave(&rchp->comp_handler_lock, flag); (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context); spin_unlock_irqrestore(&rchp->comp_handler_lock, flag); if (schp != rchp) { t4_set_cq_in_error(&schp->cq); spin_lock_irqsave(&schp->comp_handler_lock, flag); (*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context); spin_unlock_irqrestore(&schp->comp_handler_lock, flag); } return; } __flush_qp(qhp, rchp, schp); } static int rdma_fini(struct c4iw_dev *rhp, struct c4iw_qp *qhp, struct c4iw_ep *ep) { struct c4iw_rdev *rdev = &rhp->rdev; struct adapter *sc = rdev->adap; struct fw_ri_wr *wqe; int ret; struct wrqe *wr; struct socket *so = ep->com.so; struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = intotcpcb(inp); struct toepcb *toep = tp->t_toe; KASSERT(rhp == qhp->rhp && ep == qhp->ep, ("%s: EDOOFUS", __func__)); CTR5(KTR_IW_CXGBE, "%s qhp %p qid 0x%x ep %p tid %u", __func__, qhp, qhp->wq.sq.qid, ep, ep->hwtid); wr = alloc_wrqe(sizeof(*wqe), toep->ofld_txq); if (wr == NULL) return (0); wqe = wrtod(wr); memset(wqe, 0, sizeof *wqe); wqe->op_compl = cpu_to_be32(V_FW_WR_OP(FW_RI_WR) | F_FW_WR_COMPL); wqe->flowid_len16 = cpu_to_be32(V_FW_WR_FLOWID(ep->hwtid) | V_FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe, 16))); wqe->cookie = (unsigned long) &ep->com.wr_wait; wqe->u.fini.type = FW_RI_TYPE_FINI; c4iw_init_wr_wait(&ep->com.wr_wait); ret = creds(toep, inp, sizeof(*wqe)); if (ret) { free_wrqe(wr); return ret; } t4_wrq_tx(sc, wr); ret = c4iw_wait_for_reply(rdev, &ep->com.wr_wait, ep->hwtid, qhp->wq.sq.qid, ep->com.so, __func__); return ret; } static void build_rtr_msg(u8 p2p_type, struct fw_ri_init *init) { CTR2(KTR_IW_CXGBE, "%s p2p_type = %d", __func__, p2p_type); memset(&init->u, 0, sizeof init->u); switch (p2p_type) { case FW_RI_INIT_P2PTYPE_RDMA_WRITE: init->u.write.opcode = FW_RI_RDMA_WRITE_WR; init->u.write.stag_sink = cpu_to_be32(1); init->u.write.to_sink = cpu_to_be64(1); init->u.write.u.immd_src[0].op = FW_RI_DATA_IMMD; init->u.write.len16 = DIV_ROUND_UP(sizeof init->u.write + sizeof(struct fw_ri_immd), 16); break; case FW_RI_INIT_P2PTYPE_READ_REQ: init->u.write.opcode = FW_RI_RDMA_READ_WR; init->u.read.stag_src = cpu_to_be32(1); init->u.read.to_src_lo = cpu_to_be32(1); init->u.read.stag_sink = cpu_to_be32(1); init->u.read.to_sink_lo = cpu_to_be32(1); init->u.read.len16 = DIV_ROUND_UP(sizeof init->u.read, 16); break; } } static int creds(struct toepcb *toep, struct inpcb *inp, size_t wrsize) { struct ofld_tx_sdesc *txsd; CTR3(KTR_IW_CXGBE, "%s:creB %p %u", __func__, toep , wrsize); INP_WLOCK(inp); if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) != 0) { INP_WUNLOCK(inp); return (EINVAL); } txsd = &toep->txsd[toep->txsd_pidx]; txsd->tx_credits = howmany(wrsize, 16); txsd->plen = 0; KASSERT(toep->tx_credits >= txsd->tx_credits && toep->txsd_avail > 0, ("%s: not enough credits (%d)", __func__, toep->tx_credits)); toep->tx_credits -= txsd->tx_credits; if (__predict_false(++toep->txsd_pidx == toep->txsd_total)) toep->txsd_pidx = 0; toep->txsd_avail--; INP_WUNLOCK(inp); CTR5(KTR_IW_CXGBE, "%s:creE %p %u %u %u", __func__, toep , txsd->tx_credits, toep->tx_credits, toep->txsd_pidx); return (0); } static int rdma_init(struct c4iw_dev *rhp, struct c4iw_qp *qhp) { struct fw_ri_wr *wqe; int ret; struct wrqe *wr; struct c4iw_ep *ep = qhp->ep; struct c4iw_rdev *rdev = &qhp->rhp->rdev; struct adapter *sc = rdev->adap; struct socket *so = ep->com.so; struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = intotcpcb(inp); struct toepcb *toep = tp->t_toe; CTR5(KTR_IW_CXGBE, "%s qhp %p qid 0x%x ep %p tid %u", __func__, qhp, qhp->wq.sq.qid, ep, ep->hwtid); wr = alloc_wrqe(sizeof(*wqe), toep->ofld_txq); if (wr == NULL) return (0); wqe = wrtod(wr); ret = alloc_ird(rhp, qhp->attr.max_ird); if (ret) { qhp->attr.max_ird = 0; free_wrqe(wr); return ret; } memset(wqe, 0, sizeof *wqe); wqe->op_compl = cpu_to_be32( V_FW_WR_OP(FW_RI_WR) | F_FW_WR_COMPL); wqe->flowid_len16 = cpu_to_be32(V_FW_WR_FLOWID(ep->hwtid) | V_FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe, 16))); wqe->cookie = (unsigned long) &ep->com.wr_wait; wqe->u.init.type = FW_RI_TYPE_INIT; wqe->u.init.mpareqbit_p2ptype = V_FW_RI_WR_MPAREQBIT(qhp->attr.mpa_attr.initiator) | V_FW_RI_WR_P2PTYPE(qhp->attr.mpa_attr.p2p_type); wqe->u.init.mpa_attrs = FW_RI_MPA_IETF_ENABLE; if (qhp->attr.mpa_attr.recv_marker_enabled) wqe->u.init.mpa_attrs |= FW_RI_MPA_RX_MARKER_ENABLE; if (qhp->attr.mpa_attr.xmit_marker_enabled) wqe->u.init.mpa_attrs |= FW_RI_MPA_TX_MARKER_ENABLE; if (qhp->attr.mpa_attr.crc_enabled) wqe->u.init.mpa_attrs |= FW_RI_MPA_CRC_ENABLE; wqe->u.init.qp_caps = FW_RI_QP_RDMA_READ_ENABLE | FW_RI_QP_RDMA_WRITE_ENABLE | FW_RI_QP_BIND_ENABLE; if (!qhp->ibqp.uobject) wqe->u.init.qp_caps |= FW_RI_QP_FAST_REGISTER_ENABLE | FW_RI_QP_STAG0_ENABLE; wqe->u.init.nrqe = cpu_to_be16(t4_rqes_posted(&qhp->wq)); wqe->u.init.pdid = cpu_to_be32(qhp->attr.pd); wqe->u.init.qpid = cpu_to_be32(qhp->wq.sq.qid); wqe->u.init.sq_eqid = cpu_to_be32(qhp->wq.sq.qid); wqe->u.init.rq_eqid = cpu_to_be32(qhp->wq.rq.qid); wqe->u.init.scqid = cpu_to_be32(qhp->attr.scq); wqe->u.init.rcqid = cpu_to_be32(qhp->attr.rcq); wqe->u.init.ord_max = cpu_to_be32(qhp->attr.max_ord); wqe->u.init.ird_max = cpu_to_be32(qhp->attr.max_ird); wqe->u.init.iss = cpu_to_be32(ep->snd_seq); wqe->u.init.irs = cpu_to_be32(ep->rcv_seq); wqe->u.init.hwrqsize = cpu_to_be32(qhp->wq.rq.rqt_size); wqe->u.init.hwrqaddr = cpu_to_be32(qhp->wq.rq.rqt_hwaddr - sc->vres.rq.start); if (qhp->attr.mpa_attr.initiator) build_rtr_msg(qhp->attr.mpa_attr.p2p_type, &wqe->u.init); c4iw_init_wr_wait(&ep->com.wr_wait); ret = creds(toep, inp, sizeof(*wqe)); if (ret) { free_wrqe(wr); free_ird(rhp, qhp->attr.max_ird); return ret; } t4_wrq_tx(sc, wr); ret = c4iw_wait_for_reply(rdev, &ep->com.wr_wait, ep->hwtid, qhp->wq.sq.qid, ep->com.so, __func__); toep->ulp_mode = ULP_MODE_RDMA; free_ird(rhp, qhp->attr.max_ird); return ret; } int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp, enum c4iw_qp_attr_mask mask, struct c4iw_qp_attributes *attrs, int internal) { int ret = 0; struct c4iw_qp_attributes newattr = qhp->attr; int disconnect = 0; int terminate = 0; int abort = 0; int free = 0; struct c4iw_ep *ep = NULL; CTR5(KTR_IW_CXGBE, "%s qhp %p sqid 0x%x rqid 0x%x ep %p", __func__, qhp, qhp->wq.sq.qid, qhp->wq.rq.qid, qhp->ep); CTR3(KTR_IW_CXGBE, "%s state %d -> %d", __func__, qhp->attr.state, (mask & C4IW_QP_ATTR_NEXT_STATE) ? attrs->next_state : -1); mutex_lock(&qhp->mutex); /* Process attr changes if in IDLE */ if (mask & C4IW_QP_ATTR_VALID_MODIFY) { if (qhp->attr.state != C4IW_QP_STATE_IDLE) { ret = -EIO; goto out; } if (mask & C4IW_QP_ATTR_ENABLE_RDMA_READ) newattr.enable_rdma_read = attrs->enable_rdma_read; if (mask & C4IW_QP_ATTR_ENABLE_RDMA_WRITE) newattr.enable_rdma_write = attrs->enable_rdma_write; if (mask & C4IW_QP_ATTR_ENABLE_RDMA_BIND) newattr.enable_bind = attrs->enable_bind; if (mask & C4IW_QP_ATTR_MAX_ORD) { if (attrs->max_ord > c4iw_max_read_depth) { ret = -EINVAL; goto out; } newattr.max_ord = attrs->max_ord; } if (mask & C4IW_QP_ATTR_MAX_IRD) { if (attrs->max_ird > cur_max_read_depth(rhp)) { ret = -EINVAL; goto out; } newattr.max_ird = attrs->max_ird; } qhp->attr = newattr; } if (!(mask & C4IW_QP_ATTR_NEXT_STATE)) goto out; if (qhp->attr.state == attrs->next_state) goto out; + /* Return EINPROGRESS if QP is already in transition state. + * Eg: CLOSING->IDLE transition or *->ERROR transition. + * This can happen while connection is switching(due to rdma_fini) + * from iWARP/RDDP to TOE mode and any inflight RDMA RX data will + * reach TOE driver -> TCP stack -> iWARP driver. In this way + * iWARP driver keep receiving inflight RDMA RX data until socket + * is closed or aborted. And if iWARP CM is in FPDU sate, then + * it tries to put QP in TERM state and disconnects endpoint. + * But as QP is already in transition state, this event is ignored. + */ + if ((qhp->attr.state >= C4IW_QP_STATE_ERROR) && + (attrs->next_state == C4IW_QP_STATE_TERMINATE)) { + ret = -EINPROGRESS; + goto out; + } + switch (qhp->attr.state) { case C4IW_QP_STATE_IDLE: switch (attrs->next_state) { case C4IW_QP_STATE_RTS: if (!(mask & C4IW_QP_ATTR_LLP_STREAM_HANDLE)) { ret = -EINVAL; goto out; } if (!(mask & C4IW_QP_ATTR_MPA_ATTR)) { ret = -EINVAL; goto out; } qhp->attr.mpa_attr = attrs->mpa_attr; qhp->attr.llp_stream_handle = attrs->llp_stream_handle; qhp->ep = qhp->attr.llp_stream_handle; set_state(qhp, C4IW_QP_STATE_RTS); /* * Ref the endpoint here and deref when we * disassociate the endpoint from the QP. This * happens in CLOSING->IDLE transition or *->ERROR * transition. */ c4iw_get_ep(&qhp->ep->com); ret = rdma_init(rhp, qhp); if (ret) goto err; break; case C4IW_QP_STATE_ERROR: set_state(qhp, C4IW_QP_STATE_ERROR); flush_qp(qhp); break; default: ret = -EINVAL; goto out; } break; case C4IW_QP_STATE_RTS: switch (attrs->next_state) { case C4IW_QP_STATE_CLOSING: BUG_ON(atomic_read(&qhp->ep->com.kref.refcount) < 2); t4_set_wq_in_error(&qhp->wq); set_state(qhp, C4IW_QP_STATE_CLOSING); ep = qhp->ep; if (!internal) { abort = 0; disconnect = 1; c4iw_get_ep(&qhp->ep->com); } ret = rdma_fini(rhp, qhp, ep); if (ret) goto err; break; case C4IW_QP_STATE_TERMINATE: t4_set_wq_in_error(&qhp->wq); set_state(qhp, C4IW_QP_STATE_TERMINATE); qhp->attr.layer_etype = attrs->layer_etype; qhp->attr.ecode = attrs->ecode; ep = qhp->ep; if (!internal) { c4iw_get_ep(&qhp->ep->com); terminate = 1; disconnect = 1; } else { terminate = qhp->attr.send_term; ret = rdma_fini(rhp, qhp, ep); if (ret) goto err; } break; case C4IW_QP_STATE_ERROR: t4_set_wq_in_error(&qhp->wq); set_state(qhp, C4IW_QP_STATE_ERROR); if (!internal) { abort = 1; disconnect = 1; ep = qhp->ep; c4iw_get_ep(&qhp->ep->com); } goto err; break; default: ret = -EINVAL; goto out; } break; case C4IW_QP_STATE_CLOSING: /* * Allow kernel users to move to ERROR for qp draining. */ if (!internal && (qhp->ibqp.uobject || attrs->next_state != C4IW_QP_STATE_ERROR)) { ret = -EINVAL; goto out; } switch (attrs->next_state) { case C4IW_QP_STATE_IDLE: flush_qp(qhp); set_state(qhp, C4IW_QP_STATE_IDLE); qhp->attr.llp_stream_handle = NULL; c4iw_put_ep(&qhp->ep->com); qhp->ep = NULL; wake_up(&qhp->wait); break; case C4IW_QP_STATE_ERROR: goto err; default: ret = -EINVAL; goto err; } break; case C4IW_QP_STATE_ERROR: if (attrs->next_state != C4IW_QP_STATE_IDLE) { ret = -EINVAL; goto out; } if (!t4_sq_empty(&qhp->wq) || !t4_rq_empty(&qhp->wq)) { ret = -EINVAL; goto out; } set_state(qhp, C4IW_QP_STATE_IDLE); break; case C4IW_QP_STATE_TERMINATE: if (!internal) { ret = -EINVAL; goto out; } goto err; break; default: printf("%s in a bad state %d\n", __func__, qhp->attr.state); ret = -EINVAL; goto err; break; } goto out; err: CTR3(KTR_IW_CXGBE, "%s disassociating ep %p qpid 0x%x", __func__, qhp->ep, qhp->wq.sq.qid); /* disassociate the LLP connection */ qhp->attr.llp_stream_handle = NULL; if (!ep) ep = qhp->ep; qhp->ep = NULL; set_state(qhp, C4IW_QP_STATE_ERROR); free = 1; abort = 1; BUG_ON(!ep); flush_qp(qhp); wake_up(&qhp->wait); out: mutex_unlock(&qhp->mutex); if (terminate) post_terminate(qhp, NULL, internal ? GFP_ATOMIC : GFP_KERNEL); /* * If disconnect is 1, then we need to initiate a disconnect * on the EP. This can be a normal close (RTS->CLOSING) or * an abnormal close (RTS/CLOSING->ERROR). */ if (disconnect) { __c4iw_ep_disconnect(ep, abort, internal ? GFP_ATOMIC : GFP_KERNEL); c4iw_put_ep(&ep->com); } /* * If free is 1, then we've disassociated the EP from the QP * and we need to dereference the EP. */ if (free) c4iw_put_ep(&ep->com); CTR2(KTR_IW_CXGBE, "%s exit state %d", __func__, qhp->attr.state); return ret; } int c4iw_destroy_qp(struct ib_qp *ib_qp) { struct c4iw_dev *rhp; struct c4iw_qp *qhp; struct c4iw_qp_attributes attrs; CTR2(KTR_IW_CXGBE, "%s ib_qp %p", __func__, ib_qp); qhp = to_c4iw_qp(ib_qp); rhp = qhp->rhp; attrs.next_state = C4IW_QP_STATE_ERROR; if (qhp->attr.state == C4IW_QP_STATE_TERMINATE) c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); else c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 0); wait_event(qhp->wait, !qhp->ep); remove_handle(rhp, &rhp->qpidr, qhp->wq.sq.qid); free_ird(rhp, qhp->attr.max_ird); c4iw_qp_rem_ref(ib_qp); CTR3(KTR_IW_CXGBE, "%s ib_qp %p qpid 0x%0x", __func__, ib_qp, qhp->wq.sq.qid); return 0; } struct ib_qp * c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs, struct ib_udata *udata) { struct c4iw_dev *rhp; struct c4iw_qp *qhp; struct c4iw_pd *php; struct c4iw_cq *schp; struct c4iw_cq *rchp; struct c4iw_create_qp_resp uresp; unsigned int sqsize, rqsize; struct c4iw_ucontext *ucontext; int ret; struct c4iw_mm_entry *sq_key_mm = NULL, *rq_key_mm = NULL; struct c4iw_mm_entry *sq_db_key_mm = NULL, *rq_db_key_mm = NULL; CTR2(KTR_IW_CXGBE, "%s ib_pd %p", __func__, pd); if (attrs->qp_type != IB_QPT_RC) return ERR_PTR(-EINVAL); php = to_c4iw_pd(pd); rhp = php->rhp; schp = get_chp(rhp, ((struct c4iw_cq *)attrs->send_cq)->cq.cqid); rchp = get_chp(rhp, ((struct c4iw_cq *)attrs->recv_cq)->cq.cqid); if (!schp || !rchp) return ERR_PTR(-EINVAL); if (attrs->cap.max_inline_data > T4_MAX_SEND_INLINE) return ERR_PTR(-EINVAL); if (attrs->cap.max_recv_wr > rhp->rdev.hw_queue.t4_max_rq_size) return ERR_PTR(-E2BIG); rqsize = attrs->cap.max_recv_wr + 1; if (rqsize < 8) rqsize = 8; if (attrs->cap.max_send_wr > rhp->rdev.hw_queue.t4_max_sq_size) return ERR_PTR(-E2BIG); sqsize = attrs->cap.max_send_wr + 1; if (sqsize < 8) sqsize = 8; ucontext = pd->uobject ? to_c4iw_ucontext(pd->uobject->context) : NULL; qhp = kzalloc(sizeof(*qhp), GFP_KERNEL); if (!qhp) return ERR_PTR(-ENOMEM); qhp->wq.sq.size = sqsize; qhp->wq.sq.memsize = (sqsize + rhp->rdev.hw_queue.t4_eq_status_entries) * sizeof(*qhp->wq.sq.queue) + 16 * sizeof(__be64); qhp->wq.sq.flush_cidx = -1; qhp->wq.rq.size = rqsize; qhp->wq.rq.memsize = (rqsize + rhp->rdev.hw_queue.t4_eq_status_entries) * sizeof(*qhp->wq.rq.queue); if (ucontext) { qhp->wq.sq.memsize = roundup(qhp->wq.sq.memsize, PAGE_SIZE); qhp->wq.rq.memsize = roundup(qhp->wq.rq.memsize, PAGE_SIZE); } CTR5(KTR_IW_CXGBE, "%s sqsize %u sqmemsize %zu rqsize %u rqmemsize %zu", __func__, sqsize, qhp->wq.sq.memsize, rqsize, qhp->wq.rq.memsize); ret = create_qp(&rhp->rdev, &qhp->wq, &schp->cq, &rchp->cq, ucontext ? &ucontext->uctx : &rhp->rdev.uctx); if (ret) goto err1; attrs->cap.max_recv_wr = rqsize - 1; attrs->cap.max_send_wr = sqsize - 1; attrs->cap.max_inline_data = T4_MAX_SEND_INLINE; qhp->rhp = rhp; qhp->attr.pd = php->pdid; qhp->attr.scq = ((struct c4iw_cq *) attrs->send_cq)->cq.cqid; qhp->attr.rcq = ((struct c4iw_cq *) attrs->recv_cq)->cq.cqid; qhp->attr.sq_num_entries = attrs->cap.max_send_wr; qhp->attr.rq_num_entries = attrs->cap.max_recv_wr; qhp->attr.sq_max_sges = attrs->cap.max_send_sge; qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge; qhp->attr.rq_max_sges = attrs->cap.max_recv_sge; qhp->attr.state = C4IW_QP_STATE_IDLE; qhp->attr.next_state = C4IW_QP_STATE_IDLE; qhp->attr.enable_rdma_read = 1; qhp->attr.enable_rdma_write = 1; qhp->attr.enable_bind = 1; qhp->attr.max_ord = 0; qhp->attr.max_ird = 0; qhp->sq_sig_all = attrs->sq_sig_type == IB_SIGNAL_ALL_WR; spin_lock_init(&qhp->lock); mutex_init(&qhp->mutex); init_waitqueue_head(&qhp->wait); kref_init(&qhp->kref); INIT_WORK(&qhp->free_work, free_qp_work); ret = insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.sq.qid); if (ret) goto err2; if (udata) { sq_key_mm = kmalloc(sizeof(*sq_key_mm), GFP_KERNEL); if (!sq_key_mm) { ret = -ENOMEM; goto err3; } rq_key_mm = kmalloc(sizeof(*rq_key_mm), GFP_KERNEL); if (!rq_key_mm) { ret = -ENOMEM; goto err4; } sq_db_key_mm = kmalloc(sizeof(*sq_db_key_mm), GFP_KERNEL); if (!sq_db_key_mm) { ret = -ENOMEM; goto err5; } rq_db_key_mm = kmalloc(sizeof(*rq_db_key_mm), GFP_KERNEL); if (!rq_db_key_mm) { ret = -ENOMEM; goto err6; } uresp.flags = 0; uresp.qid_mask = rhp->rdev.qpmask; uresp.sqid = qhp->wq.sq.qid; uresp.sq_size = qhp->wq.sq.size; uresp.sq_memsize = qhp->wq.sq.memsize; uresp.rqid = qhp->wq.rq.qid; uresp.rq_size = qhp->wq.rq.size; uresp.rq_memsize = qhp->wq.rq.memsize; spin_lock(&ucontext->mmap_lock); uresp.ma_sync_key = 0; uresp.sq_key = ucontext->key; ucontext->key += PAGE_SIZE; uresp.rq_key = ucontext->key; ucontext->key += PAGE_SIZE; uresp.sq_db_gts_key = ucontext->key; ucontext->key += PAGE_SIZE; uresp.rq_db_gts_key = ucontext->key; ucontext->key += PAGE_SIZE; spin_unlock(&ucontext->mmap_lock); ret = ib_copy_to_udata(udata, &uresp, sizeof uresp); if (ret) goto err7; sq_key_mm->key = uresp.sq_key; sq_key_mm->addr = qhp->wq.sq.phys_addr; sq_key_mm->len = PAGE_ALIGN(qhp->wq.sq.memsize); CTR4(KTR_IW_CXGBE, "%s sq_key_mm %x, %x, %d", __func__, sq_key_mm->key, sq_key_mm->addr, sq_key_mm->len); insert_mmap(ucontext, sq_key_mm); rq_key_mm->key = uresp.rq_key; rq_key_mm->addr = qhp->wq.rq.phys_addr; rq_key_mm->len = PAGE_ALIGN(qhp->wq.rq.memsize); CTR4(KTR_IW_CXGBE, "%s rq_key_mm %x, %x, %d", __func__, rq_key_mm->key, rq_key_mm->addr, rq_key_mm->len); insert_mmap(ucontext, rq_key_mm); sq_db_key_mm->key = uresp.sq_db_gts_key; sq_db_key_mm->addr = (u64)qhp->wq.sq.bar2_pa; sq_db_key_mm->len = PAGE_SIZE; CTR4(KTR_IW_CXGBE, "%s sq_db_key_mm %x, %x, %d", __func__, sq_db_key_mm->key, sq_db_key_mm->addr, sq_db_key_mm->len); insert_mmap(ucontext, sq_db_key_mm); rq_db_key_mm->key = uresp.rq_db_gts_key; rq_db_key_mm->addr = (u64)qhp->wq.rq.bar2_pa; rq_db_key_mm->len = PAGE_SIZE; CTR4(KTR_IW_CXGBE, "%s rq_db_key_mm %x, %x, %d", __func__, rq_db_key_mm->key, rq_db_key_mm->addr, rq_db_key_mm->len); insert_mmap(ucontext, rq_db_key_mm); c4iw_get_ucontext(ucontext); qhp->ucontext = ucontext; } qhp->ibqp.qp_num = qhp->wq.sq.qid; init_timer(&(qhp->timer)); CTR5(KTR_IW_CXGBE, "%s sq id %u size %u memsize %zu num_entries %u\n", __func__, qhp->wq.sq.qid, qhp->wq.sq.size, qhp->wq.sq.memsize, attrs->cap.max_send_wr); CTR5(KTR_IW_CXGBE, "%s rq id %u size %u memsize %zu num_entries %u\n", __func__, qhp->wq.rq.qid, qhp->wq.rq.size, qhp->wq.rq.memsize, attrs->cap.max_recv_wr); return &qhp->ibqp; err7: kfree(rq_db_key_mm); err6: kfree(sq_db_key_mm); err5: kfree(rq_key_mm); err4: kfree(sq_key_mm); err3: remove_handle(rhp, &rhp->qpidr, qhp->wq.sq.qid); err2: destroy_qp(&rhp->rdev, &qhp->wq, ucontext ? &ucontext->uctx : &rhp->rdev.uctx); err1: kfree(qhp); return ERR_PTR(ret); } int c4iw_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask, struct ib_udata *udata) { struct c4iw_dev *rhp; struct c4iw_qp *qhp; enum c4iw_qp_attr_mask mask = 0; struct c4iw_qp_attributes attrs; CTR2(KTR_IW_CXGBE, "%s ib_qp %p", __func__, ibqp); /* iwarp does not support the RTR state */ if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR)) attr_mask &= ~IB_QP_STATE; /* Make sure we still have something left to do */ if (!attr_mask) return 0; memset(&attrs, 0, sizeof attrs); qhp = to_c4iw_qp(ibqp); rhp = qhp->rhp; attrs.next_state = c4iw_convert_state(attr->qp_state); attrs.enable_rdma_read = (attr->qp_access_flags & IB_ACCESS_REMOTE_READ) ? 1 : 0; attrs.enable_rdma_write = (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE) ? 1 : 0; attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0; mask |= (attr_mask & IB_QP_STATE) ? C4IW_QP_ATTR_NEXT_STATE : 0; mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ? (C4IW_QP_ATTR_ENABLE_RDMA_READ | C4IW_QP_ATTR_ENABLE_RDMA_WRITE | C4IW_QP_ATTR_ENABLE_RDMA_BIND) : 0; return c4iw_modify_qp(rhp, qhp, mask, &attrs, 0); } struct ib_qp *c4iw_get_qp(struct ib_device *dev, int qpn) { CTR3(KTR_IW_CXGBE, "%s ib_dev %p qpn 0x%x", __func__, dev, qpn); return (struct ib_qp *)get_qhp(to_c4iw_dev(dev), qpn); } int c4iw_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask, struct ib_qp_init_attr *init_attr) { struct c4iw_qp *qhp = to_c4iw_qp(ibqp); memset(attr, 0, sizeof *attr); memset(init_attr, 0, sizeof *init_attr); attr->qp_state = to_ib_qp_state(qhp->attr.state); init_attr->cap.max_send_wr = qhp->attr.sq_num_entries; init_attr->cap.max_recv_wr = qhp->attr.rq_num_entries; init_attr->cap.max_send_sge = qhp->attr.sq_max_sges; init_attr->cap.max_recv_sge = qhp->attr.sq_max_sges; init_attr->cap.max_inline_data = T4_MAX_SEND_INLINE; init_attr->sq_sig_type = qhp->sq_sig_all ? IB_SIGNAL_ALL_WR : 0; return 0; } #endif