Index: stable/10/sys/dev/cxgbe/iw_cxgbe/cm.c =================================================================== --- stable/10/sys/dev/cxgbe/iw_cxgbe/cm.c (revision 283855) +++ stable/10/sys/dev/cxgbe/iw_cxgbe/cm.c (revision 283856) @@ -1,2455 +1,2459 @@ /* * 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; #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 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(timeout_list); static spinlock_t timeout_lock; static void process_req(struct work_struct *ctx); static void start_ep_timer(struct c4iw_ep *ep); static void stop_ep_timer(struct c4iw_ep *ep); static int set_tcpinfo(struct c4iw_ep *ep); static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc); static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate); static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate); static void *alloc_ep(int size, gfp_t flags); void __free_ep(struct c4iw_ep_common *epc); static struct rtentry * find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port, __be16 peer_port, u8 tos); static int close_socket(struct c4iw_ep_common *epc, int close); static int shutdown_socket(struct c4iw_ep_common *epc); static void abort_socket(struct c4iw_ep *ep); static void 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); +static void close_complete_upcall(struct c4iw_ep *ep, int status); static int abort_connection(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 void connect_request_upcall(struct c4iw_ep *ep); static void established_upcall(struct c4iw_ep *ep); static void process_mpa_reply(struct c4iw_ep *ep); static void 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 init_sock(struct c4iw_ep_common *epc); static void process_data(struct c4iw_ep *ep); static void process_connected(struct c4iw_ep *ep); static struct socket * dequeue_socket(struct socket *head, struct sockaddr_in **remote, struct c4iw_ep *child_ep); static void process_newconn(struct c4iw_ep *parent_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); #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) \ do { \ CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \ __func__, __LINE__, (ep)); \ stop_ep_timer(ep); \ } 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 process_req(struct work_struct *ctx) { struct c4iw_ep_common *epc; spin_lock(&req_lock); while (!TAILQ_EMPTY(&req_list)) { epc = TAILQ_FIRST(&req_list); TAILQ_REMOVE(&req_list, epc, entry); epc->entry.tqe_prev = NULL; spin_unlock(&req_lock); if (epc->so) process_socket_event((struct c4iw_ep *)epc); c4iw_put_ep(epc); spin_lock(&req_lock); } spin_unlock(&req_lock); } /* * 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 struct rtentry * find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port, __be16 peer_port, u8 tos) { struct route iproute; struct sockaddr_in *dst = (struct sockaddr_in *)&iproute.ro_dst; CTR5(KTR_IW_CXGBE, "%s:frtB %x, %x, %d, %d", __func__, local_ip, peer_ip, ntohs(local_port), ntohs(peer_port)); bzero(&iproute, sizeof iproute); dst->sin_family = AF_INET; dst->sin_len = sizeof *dst; dst->sin_addr.s_addr = peer_ip; rtalloc(&iproute); CTR2(KTR_IW_CXGBE, "%s:frtE %p", __func__, (uint64_t)iproute.ro_rt); return iproute.ro_rt; } static int close_socket(struct c4iw_ep_common *epc, int close) { struct socket *so = epc->so; int rc; CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc, so, states[epc->state]); SOCK_LOCK(so); soupcall_clear(so, SO_RCV); SOCK_UNLOCK(so); if (close) rc = soclose(so); else rc = soshutdown(so, SHUT_WR | SHUT_RD); epc->so = NULL; return (rc); } static int shutdown_socket(struct c4iw_ep_common *epc) { CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc->so, epc, states[epc->state]); return (soshutdown(epc->so, SHUT_WR)); } static void abort_socket(struct c4iw_ep *ep) { struct sockopt sopt; int rc; struct linger l; CTR4(KTR_IW_CXGBE, "%s ep %p so %p state %s", __func__, ep, ep->com.so, states[ep->com.state]); 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(ep->com.so, &sopt); if (rc) { log(LOG_ERR, "%s: can't set linger to 0, no RST! err %d\n", __func__, rc); } } static void process_peer_close(struct c4iw_ep *ep) { struct c4iw_qp_attributes attrs; 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]); mutex_lock(&ep->com.mutex); switch (ep->com.state) { case MPA_REQ_WAIT: CTR2(KTR_IW_CXGBE, "%s:ppc1 %p MPA_REQ_WAIT CLOSING", __func__, ep); __state_set(&ep->com, CLOSING); break; case MPA_REQ_SENT: CTR2(KTR_IW_CXGBE, "%s:ppc2 %p MPA_REQ_SENT CLOSING", __func__, ep); __state_set(&ep->com, DEAD); connect_reply_upcall(ep, -ECONNABORTED); disconnect = 0; STOP_EP_TIMER(ep); close_socket(&ep->com, 0); ep->com.cm_id->rem_ref(ep->com.cm_id); ep->com.cm_id = NULL; ep->com.qp = NULL; 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); __state_set(&ep->com, CLOSING); c4iw_get_ep(&ep->com); break; case MPA_REP_SENT: CTR2(KTR_IW_CXGBE, "%s:ppc4 %p MPA_REP_SENT CLOSING", __func__, ep); __state_set(&ep->com, CLOSING); break; case FPDU_MODE: CTR2(KTR_IW_CXGBE, "%s:ppc5 %p FPDU_MODE CLOSING", __func__, ep); START_EP_TIMER(ep); __state_set(&ep->com, 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); __state_set(&ep->com, 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, 0); - close_complete_upcall(ep); + close_complete_upcall(ep, 0); __state_set(&ep->com, 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; } mutex_unlock(&ep->com.mutex); 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; int ret; int state; state = state_read(&ep->com); 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); 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: /* * We're gonna mark this puppy DEAD, but keep * the reference on it until the ULP accepts or * rejects the CR. */ c4iw_get_ep(&ep->com); 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) { CTR2(KTR_IW_CXGBE, "%s:pce1 %p", __func__, ep); - close_socket(&ep->com, 0); + close_socket(&ep->com, 1); state_set(&ep->com, 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; 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 */ mutex_lock(&ep->com.mutex); switch (ep->com.state) { case CLOSING: CTR2(KTR_IW_CXGBE, "%s:pcc1 %p CLOSING MORIBUND", __func__, ep); __state_set(&ep->com, 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); } if (ep->parent_ep) { CTR2(KTR_IW_CXGBE, "%s:pcc3 %p", __func__, ep); close_socket(&ep->com, 1); } else { CTR2(KTR_IW_CXGBE, "%s:pcc4 %p", __func__, ep); close_socket(&ep->com, 0); } - close_complete_upcall(ep); + close_complete_upcall(ep, 0); __state_set(&ep->com, DEAD); release = 1; break; case ABORTING: CTR2(KTR_IW_CXGBE, "%s:pcc5 %p ABORTING", __func__, ep); break; case DEAD: default: CTR2(KTR_IW_CXGBE, "%s:pcc6 %p DEAD", __func__, ep); panic("%s:pcc6 %p DEAD", __func__, ep); break; } mutex_unlock(&ep->com.mutex); if (release) { CTR2(KTR_IW_CXGBE, "%s:pcc7 %p", __func__, ep); c4iw_put_ep(&ep->com); } CTR2(KTR_IW_CXGBE, "%s:pccE %p", __func__, ep); return; } static void init_sock(struct c4iw_ep_common *epc) { int rc; struct sockopt sopt; struct socket *so = epc->so; int on = 1; SOCK_LOCK(so); soupcall_set(so, SO_RCV, c4iw_so_upcall, epc); so->so_state |= SS_NBIO; SOCK_UNLOCK(so); 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 process_data(struct c4iw_ep *ep) { struct sockaddr_in *local, *remote; CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sb_cc %d", __func__, ep->com.so, ep, states[ep->com.state], ep->com.so->so_rcv.sb_cc); switch (state_read(&ep->com)) { case MPA_REQ_SENT: process_mpa_reply(ep); break; case MPA_REQ_WAIT: in_getsockaddr(ep->com.so, (struct sockaddr **)&local); in_getpeeraddr(ep->com.so, (struct sockaddr **)&remote); ep->com.local_addr = *local; ep->com.remote_addr = *remote; free(local, M_SONAME); free(remote, M_SONAME); process_mpa_request(ep); break; default: if (ep->com.so->so_rcv.sb_cc) log(LOG_ERR, "%s: Unexpected streaming data. " "ep %p, state %d, so %p, so_state 0x%x, sb_cc %u\n", __func__, ep, state_read(&ep->com), ep->com.so, ep->com.so->so_state, ep->com.so->so_rcv.sb_cc); break; } } static void process_connected(struct c4iw_ep *ep) { if ((ep->com.so->so_state & SS_ISCONNECTED) && !ep->com.so->so_error) send_mpa_req(ep); else { connect_reply_upcall(ep, -ep->com.so->so_error); close_socket(&ep->com, 0); state_set(&ep->com, DEAD); c4iw_put_ep(&ep->com); } } static struct socket * dequeue_socket(struct socket *head, struct sockaddr_in **remote, struct c4iw_ep *child_ep) { struct socket *so; ACCEPT_LOCK(); so = TAILQ_FIRST(&head->so_comp); if (!so) { ACCEPT_UNLOCK(); return (NULL); } TAILQ_REMOVE(&head->so_comp, so, so_list); head->so_qlen--; SOCK_LOCK(so); so->so_qstate &= ~SQ_COMP; so->so_head = NULL; soref(so); soupcall_set(so, SO_RCV, c4iw_so_upcall, child_ep); so->so_state |= SS_NBIO; SOCK_UNLOCK(so); ACCEPT_UNLOCK(); soaccept(so, (struct sockaddr **)remote); return (so); } static void process_newconn(struct c4iw_ep *parent_ep) { struct socket *child_so; struct c4iw_ep *child_ep; struct sockaddr_in *remote; child_ep = alloc_ep(sizeof(*child_ep), M_NOWAIT); if (!child_ep) { CTR3(KTR_IW_CXGBE, "%s: parent so %p, parent ep %p, ENOMEM", __func__, parent_ep->com.so, parent_ep); log(LOG_ERR, "%s: failed to allocate ep entry\n", __func__); return; } child_so = dequeue_socket(parent_ep->com.so, &remote, child_ep); if (!child_so) { CTR4(KTR_IW_CXGBE, "%s: parent so %p, parent ep %p, child ep %p, dequeue err", __func__, parent_ep->com.so, parent_ep, child_ep); log(LOG_ERR, "%s: failed to dequeue child socket\n", __func__); __free_ep(&child_ep->com); return; } CTR5(KTR_IW_CXGBE, "%s: parent so %p, parent ep %p, child so %p, child ep %p", __func__, parent_ep->com.so, parent_ep, child_so, child_ep); child_ep->com.local_addr = parent_ep->com.local_addr; child_ep->com.remote_addr = *remote; child_ep->com.dev = parent_ep->com.dev; child_ep->com.so = child_so; child_ep->com.cm_id = NULL; child_ep->com.thread = parent_ep->com.thread; child_ep->parent_ep = parent_ep; free(remote, M_SONAME); c4iw_get_ep(&parent_ep->com); child_ep->parent_ep = parent_ep; init_timer(&child_ep->timer); state_set(&child_ep->com, MPA_REQ_WAIT); START_EP_TIMER(child_ep); /* maybe the request has already been queued up on the socket... */ process_mpa_request(child_ep); } static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag) { struct c4iw_ep *ep = arg; spin_lock(&req_lock); 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); if (ep && ep->com.so && !ep->com.entry.tqe_prev) { KASSERT(ep->com.so == so, ("%s: XXX review.", __func__)); c4iw_get_ep(&ep->com); TAILQ_INSERT_TAIL(&req_list, &ep->com, entry); queue_work(c4iw_taskq, &c4iw_task); } spin_unlock(&req_lock); return (SU_OK); } static void process_socket_event(struct c4iw_ep *ep) { int state = state_read(&ep->com); struct socket *so = ep->com.so; 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) { process_newconn(ep); 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); return; } /* close complete */ if (so->so_state & SS_ISDISCONNECTED) { process_close_complete(ep); return; } /* rx data */ process_data(ep); } SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters"); int db_delay_usecs = 1; TUNABLE_INT("hw.iw_cxgbe.db_delay_usecs", &db_delay_usecs); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_delay_usecs, CTLFLAG_RW, &db_delay_usecs, 0, "Usecs to delay awaiting db fifo to drain"); static int dack_mode = 1; TUNABLE_INT("hw.iw_cxgbe.dack_mode", &dack_mode); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RW, &dack_mode, 0, "Delayed ack mode (default = 1)"); int c4iw_max_read_depth = 8; TUNABLE_INT("hw.iw_cxgbe.c4iw_max_read_depth", &c4iw_max_read_depth); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RW, &c4iw_max_read_depth, 0, "Per-connection max ORD/IRD (default = 8)"); static int enable_tcp_timestamps; TUNABLE_INT("hw.iw_cxgbe.enable_tcp_timestamps", &enable_tcp_timestamps); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RW, &enable_tcp_timestamps, 0, "Enable tcp timestamps (default = 0)"); static int enable_tcp_sack; TUNABLE_INT("hw.iw_cxgbe.enable_tcp_sack", &enable_tcp_sack); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RW, &enable_tcp_sack, 0, "Enable tcp SACK (default = 0)"); static int enable_tcp_window_scaling = 1; TUNABLE_INT("hw.iw_cxgbe.enable_tcp_window_scaling", &enable_tcp_window_scaling); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RW, &enable_tcp_window_scaling, 0, "Enable tcp window scaling (default = 1)"); int c4iw_debug = 1; TUNABLE_INT("hw.iw_cxgbe.c4iw_debug", &c4iw_debug); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RW, &c4iw_debug, 0, "Enable debug logging (default = 0)"); static int peer2peer; TUNABLE_INT("hw.iw_cxgbe.peer2peer", &peer2peer); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RW, &peer2peer, 0, "Support peer2peer ULPs (default = 0)"); static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; TUNABLE_INT("hw.iw_cxgbe.p2p_type", &p2p_type); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RW, &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; TUNABLE_INT("hw.iw_cxgbe.ep_timeout_secs", &ep_timeout_secs); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RW, &ep_timeout_secs, 0, "CM Endpoint operation timeout in seconds (default = 60)"); static int mpa_rev = 1; TUNABLE_INT("hw.iw_cxgbe.mpa_rev", &mpa_rev); #ifdef IW_CM_MPAV2 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RW, &mpa_rev, 0, "MPA Revision, 0 supports amso1100, 1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)"); #else SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RW, &mpa_rev, 0, "MPA Revision, 0 supports amso1100, 1 is RFC0544 spec compliant (default = 1)"); #endif static int markers_enabled; TUNABLE_INT("hw.iw_cxgbe.markers_enabled", &markers_enabled); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RW, &markers_enabled, 0, "Enable MPA MARKERS (default(0) = disabled)"); static int crc_enabled = 1; TUNABLE_INT("hw.iw_cxgbe.crc_enabled", &crc_enabled); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RW, &crc_enabled, 0, "Enable MPA CRC (default(1) = enabled)"); static int rcv_win = 256 * 1024; TUNABLE_INT("hw.iw_cxgbe.rcv_win", &rcv_win); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RW, &rcv_win, 0, "TCP receive window in bytes (default = 256KB)"); static int snd_win = 128 * 1024; TUNABLE_INT("hw.iw_cxgbe.snd_win", &snd_win); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RW, &snd_win, 0, "TCP send window in bytes (default = 128KB)"); int db_fc_threshold = 2000; TUNABLE_INT("hw.iw_cxgbe.db_fc_threshold", &db_fc_threshold); SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_fc_threshold, CTLFLAG_RW, &db_fc_threshold, 0, "QP count/threshold that triggers automatic"); 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 void 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); } } static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc) { enum c4iw_ep_state state; mutex_lock(&epc->mutex); state = epc->state; mutex_unlock(&epc->mutex); return (state); } static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) { epc->state = new; } static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) { mutex_lock(&epc->mutex); __state_set(epc, new); mutex_unlock(&epc->mutex); } 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 __free_ep(struct c4iw_ep_common *epc) { CTR2(KTR_IW_CXGBE, "%s:feB %p", __func__, epc); KASSERT(!epc->so, ("%s warning ep->so %p \n", __func__, epc->so)); KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list!\n", __func__, epc)); free(epc, M_DEVBUF); CTR2(KTR_IW_CXGBE, "%s:feE %p", __func__, 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->so, ("%s ep->so %p", __func__, epc->so)); KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list", __func__, epc)); 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 void 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; 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) { failed: connect_reply_upcall(ep, -ENOMEM); return; } 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(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) { free(mpa, M_CXGBE); goto failed; } 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) goto failed; START_EP_TIMER(ep); state_set(&ep->com, MPA_REQ_SENT); ep->mpa_attr.initiator = 1; } 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(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); state_set(&ep->com, 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) +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); ep->com.cm_id->rem_ref(ep->com.cm_id); ep->com.cm_id = NULL; ep->com.qp = NULL; set_bit(CLOSE_UPCALL, &ep->com.history); } CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep); } static int abort_connection(struct c4iw_ep *ep) { int err; CTR2(KTR_IW_CXGBE, "%s:abB %p", __func__, ep); - close_complete_upcall(ep); + close_complete_upcall(ep, -ECONNRESET); state_set(&ep->com, ABORTING); abort_socket(ep); err = close_socket(&ep->com, 0); set_bit(ABORT_CONN, &ep->com.history); CTR2(KTR_IW_CXGBE, "%s:abE %p", __func__, ep); return err; } 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); ep->com.cm_id->rem_ref(ep->com.cm_id); ep->com.cm_id = NULL; ep->com.qp = NULL; 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", __func__, ep, status); memset(&event, 0, sizeof(event)); event.event = IW_CM_EVENT_CONNECT_REPLY; event.status = (status ==-ECONNABORTED)?-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.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.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); ep->com.cm_id->rem_ref(ep->com.cm_id); ep->com.cm_id = NULL; ep->com.qp = NULL; } CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep); } static void connect_request_upcall(struct c4iw_ep *ep) { struct iw_cm_event event; 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; event.so = ep->com.so; if (!ep->tried_with_mpa_v1) { /* this means MPA_v2 is used */ #ifdef IW_CM_MPAV2 event.ord = ep->ord; event.ird = ep->ird; #endif 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 */ #ifdef IW_CM_MPAV2 event.ord = c4iw_max_read_depth; event.ird = c4iw_max_read_depth; #endif event.private_data_len = ep->plen; event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); } c4iw_get_ep(&ep->com); ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, &event); set_bit(CONNREQ_UPCALL, &ep->com.history); c4iw_put_ep(&ep->parent_ep->com); } 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; #ifdef IW_CM_MPAV2 event.ird = ep->ird; event.ord = ep->ord; #endif 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); } static void 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; enum c4iw_qp_attr_mask mask; int err; struct mbuf *top, *m; int flags = MSG_DONTWAIT; struct uio uio; CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep); /* * Stop mpa timer. If it expired, then the state has * changed and we bail since ep_timeout already aborted * the connection. */ STOP_EP_TIMER(ep); if (state_read(&ep->com) != MPA_REQ_SENT) return; 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; } 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; } /* * 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; 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; } if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep); err = -EPROTO; goto err; } 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; } /* * 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); err = -EPROTO; goto err; } 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; } if (mpa->flags & MPA_REJECT) { CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep); err = -ECONNREFUSED; goto err; } /* * If we get here we have accumulated the entire mpa * start reply message including private data. And * the MPA header is valid. */ state_set(&ep->com, 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) || (ep->ord > resp_ird)) { CTR2(KTR_IW_CXGBE, "%s:pmre %p", __func__, ep); err = -ENOMEM; ep->ird = resp_ord; ep->ord = resp_ird; insuff_ird = 1; } 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__); 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; err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 0); err = -ENOMEM; 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; err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 0); err = -ENOMEM; goto out; } goto out; err: state_set(&ep->com, ABORTING); abort_connection(ep); out: connect_reply_upcall(ep, err); CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep); return; } static void process_mpa_request(struct c4iw_ep *ep) { struct mpa_message *mpa; u16 plen; int flags = MSG_DONTWAIT; int rc; struct iovec iov; struct uio uio; enum c4iw_ep_state state = state_read(&ep->com); CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]); if (state != MPA_REQ_WAIT) return; 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; else if (rc) { abort: STOP_EP_TIMER(ep); abort_connection(ep); return; } 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; 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 abort; } if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) goto abort; /* * Fail if there's too much private data. */ plen = ntohs(mpa->private_data_size); if (plen > MPA_MAX_PRIVATE_DATA) goto abort; /* * If plen does not account for pkt size */ if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) goto abort; ep->plen = (u8) plen; /* * If we don't have all the pdata yet, then bail. */ if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) return; /* * 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) { struct mpa_v2_conn_params *mpa_v2_params; u16 ird, ord; mpa_v2_params = (void *)&ep->mpa_pkt[sizeof(*mpa)]; ird = ntohs(mpa_v2_params->ird); ord = ntohs(mpa_v2_params->ord); ep->ird = ird & MPA_V2_IRD_ORD_MASK; ep->ord = ord & MPA_V2_IRD_ORD_MASK; if (ird & MPA_V2_PEER2PEER_MODEL && peer2peer) { if (ord & MPA_V2_RDMA_WRITE_RTR) { ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_RDMA_WRITE; } else if (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 abort; 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); state_set(&ep->com, MPA_REQ_RCVD); STOP_EP_TIMER(ep); /* drive upcall */ mutex_lock(&ep->parent_ep->com.mutex); if (ep->parent_ep->com.state != DEAD) connect_request_upcall(ep); else abort_connection(ep); mutex_unlock(&ep->parent_ep->com.mutex); } /* * 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); CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep); if (state_read(&ep->com) == DEAD) { CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep); c4iw_put_ep(&ep->com); return -ECONNRESET; } set_bit(ULP_REJECT, &ep->com.history); BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD); if (mpa_rev == 0) { CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep); abort_connection(ep); } else { CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep); err = send_mpa_reject(ep, pdata, pdata_len); err = soshutdown(ep->com.so, 3); } c4iw_put_ep(&ep->com); CTR2(KTR_IW_CXGBE, "%s:crc4 %p", __func__, ep); 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; 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); CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep); if (state_read(&ep->com) == DEAD) { CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep); err = -ECONNRESET; goto err; } BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD); 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); abort_connection(ep); err = -EINVAL; goto err; } 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) { CTR2(KTR_IW_CXGBE, "%s:cac4 %p", __func__, ep); ep->ird = conn_param->ird; ep->ord = conn_param->ord; send_mpa_reject(ep, conn_param->private_data, conn_param->private_data_len); abort_connection(ep); err = -ENOMEM; goto err; } if (conn_param->ird > ep->ord) { CTR2(KTR_IW_CXGBE, "%s:cac5 %p", __func__, ep); if (!ep->ord) { CTR2(KTR_IW_CXGBE, "%s:cac6 %p", __func__, ep); conn_param->ird = 1; } else { CTR2(KTR_IW_CXGBE, "%s:cac7 %p", __func__, ep); abort_connection(ep); err = -ENOMEM; goto err; } } } ep->ird = conn_param->ird; ep->ord = conn_param->ord; if (ep->mpa_attr.version != 2) { CTR2(KTR_IW_CXGBE, "%s:cac8 %p", __func__, ep); if (peer2peer && ep->ird == 0) { CTR2(KTR_IW_CXGBE, "%s:cac9 %p", __func__, ep); ep->ird = 1; } } cm_id->add_ref(cm_id); ep->com.cm_id = cm_id; ep->com.qp = qp; //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) { CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep); goto err1; } err = send_mpa_reply(ep, conn_param->private_data, conn_param->private_data_len); if (err) { CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep); goto err1; } state_set(&ep->com, FPDU_MODE); established_upcall(ep); c4iw_put_ep(&ep->com); CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep); return 0; err1: ep->com.cm_id = NULL; ep->com.qp = NULL; cm_id->rem_ref(cm_id); err: c4iw_put_ep(&ep->com); CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep); return err; } 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 rtentry *rt; struct toedev *tdev; 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), M_NOWAIT); if (!ep) { CTR2(KTR_IW_CXGBE, "%s:cc2 %p", __func__, cm_id); printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__); err = -ENOMEM; goto out; } 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; } cm_id->add_ref(cm_id); ep->com.dev = dev; ep->com.cm_id = cm_id; 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 fail2; } ep->com.thread = curthread; ep->com.so = cm_id->so; init_sock(&ep->com); /* find a route */ rt = find_route( cm_id->local_addr.sin_addr.s_addr, cm_id->remote_addr.sin_addr.s_addr, cm_id->local_addr.sin_port, cm_id->remote_addr.sin_port, 0); if (!rt) { CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep); printk(KERN_ERR MOD "%s - cannot find route.\n", __func__); err = -EHOSTUNREACH; - goto fail3; + goto fail2; } + if (!(rt->rt_ifp->if_capenable & IFCAP_TOE)) { - if (!(rt->rt_ifp->if_flags & IFCAP_TOE)) { - CTR2(KTR_IW_CXGBE, "%s:cc8 %p", __func__, ep); printf("%s - interface not TOE capable.\n", __func__); + close_socket(&ep->com, 0); + err = -ENOPROTOOPT; goto fail3; } tdev = TOEDEV(rt->rt_ifp); if (tdev == NULL) { CTR2(KTR_IW_CXGBE, "%s:cc9 %p", __func__, ep); printf("%s - No toedev for interface.\n", __func__); goto fail3; } RTFREE(rt); state_set(&ep->com, CONNECTING); ep->tos = 0; ep->com.local_addr = cm_id->local_addr; ep->com.remote_addr = cm_id->remote_addr; err = soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr, ep->com.thread); if (!err) { - CTR2(KTR_IW_CXGBE, "%s:cca %p", __func__, ep); goto out; + } else { + close_socket(&ep->com, 0); + goto fail2; } fail3: CTR2(KTR_IW_CXGBE, "%s:ccb %p", __func__, ep); RTFREE(rt); fail2: cm_id->rem_ref(cm_id); c4iw_put_ep(&ep->com); out: CTR2(KTR_IW_CXGBE, "%s:ccE %p", __func__, ep); return err; } /* * iwcm->create_listen. Returns -errno on failure. */ int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) { int rc; struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); struct c4iw_listen_ep *ep; struct socket *so = cm_id->so; ep = alloc_ep(sizeof(*ep), GFP_KERNEL); CTR5(KTR_IW_CXGBE, "%s: cm_id %p, lso %p, ep %p, inp %p", __func__, cm_id, so, ep, so->so_pcb); if (ep == NULL) { log(LOG_ERR, "%s: failed to alloc memory for endpoint\n", __func__); rc = ENOMEM; goto failed; } cm_id->add_ref(cm_id); ep->com.cm_id = cm_id; ep->com.dev = dev; ep->backlog = backlog; ep->com.local_addr = cm_id->local_addr; ep->com.thread = curthread; state_set(&ep->com, LISTEN); ep->com.so = so; init_sock(&ep->com); rc = solisten(so, ep->backlog, ep->com.thread); if (rc != 0) { log(LOG_ERR, "%s: failed to start listener: %d\n", __func__, rc); close_socket(&ep->com, 0); cm_id->rem_ref(cm_id); c4iw_put_ep(&ep->com); goto failed; } cm_id->provider_data = ep; return (0); failed: CTR3(KTR_IW_CXGBE, "%s: cm_id %p, FAILED (%d)", __func__, cm_id, rc); return (-rc); } int c4iw_destroy_listen(struct iw_cm_id *cm_id) { int rc; struct c4iw_listen_ep *ep = to_listen_ep(cm_id); CTR4(KTR_IW_CXGBE, "%s: cm_id %p, so %p, inp %p", __func__, cm_id, cm_id->so, cm_id->so->so_pcb); state_set(&ep->com, DEAD); rc = close_socket(&ep->com, 0); cm_id->rem_ref(cm_id); c4iw_put_ep(&ep->com); return (rc); } 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; mutex_lock(&ep->com.mutex); 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); + close_complete_upcall(ep, -EIO); ep->com.state = DEAD; } CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep, states[ep->com.state]); 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; } mutex_unlock(&ep->com.mutex); 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); ret = abort_connection(ep); } else { CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep); set_bit(EP_DISC_CLOSE, &ep->com.history); if (!ep->parent_ep) __state_set(&ep->com, MORIBUND); ret = shutdown_socket(&ep->com); } if (ret) { fatal = 1; } } if (fatal) { release_ep_resources(ep); CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep); } 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; int kickit = 0; CTR2(KTR_IW_CXGBE, "%s:etB %p", __func__, ep); spin_lock(&timeout_lock); if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { list_add_tail(&ep->entry, &timeout_list); kickit = 1; } spin_unlock(&timeout_lock); if (kickit) { CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep); queue_work(c4iw_taskq, &c4iw_task); } CTR2(KTR_IW_CXGBE, "%s:etE %p", __func__, ep); } 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 t4_cqe cqe =*(const struct t4_cqe *)(&rpl[0]); CTR2(KTR_IW_CXGBE, "%s rpl %p", __func__, rpl); c4iw_ev_dispatch(sc->iwarp_softc, &cqe); return (0); } 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 *rpl = (const void *)(rss + 1); unsigned int tid = GET_TID(rpl); struct c4iw_qp_attributes attrs; struct toepcb *toep = lookup_tid(sc, tid); struct socket *so = inp_inpcbtosocket(toep->inp); struct c4iw_ep *ep = so->so_rcv.sb_upcallarg; 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", tid, 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", tid); CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep); return 0; } void c4iw_cm_init_cpl(struct adapter *sc) { t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, terminate); t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, fw6_wr_rpl); t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, fw6_cqe_handler); t4_register_an_handler(sc, c4iw_ev_handler); } void c4iw_cm_term_cpl(struct adapter *sc) { t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, NULL); t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, NULL); t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, NULL); } int __init c4iw_cm_init(void) { TAILQ_INIT(&req_list); spin_lock_init(&req_lock); INIT_LIST_HEAD(&timeout_list); spin_lock_init(&timeout_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(&timeout_list)); flush_workqueue(c4iw_taskq); destroy_workqueue(c4iw_taskq); } #endif Index: stable/10/sys/dev/cxgbe/iw_cxgbe/qp.c =================================================================== --- stable/10/sys/dev/cxgbe/iw_cxgbe/qp.c (revision 283855) +++ stable/10/sys/dev/cxgbe/iw_cxgbe/qp.c (revision 283856) @@ -1,1706 +1,1706 @@ /* * 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; #include #include "offload.h" #include "tom/t4_tom.h" #include "iw_cxgbe.h" #include "user.h" extern int db_delay_usecs; extern int db_fc_threshold; static void creds(struct toepcb *toep, size_t wrsize); 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 void dealloc_host_sq(struct c4iw_rdev *rdev, struct t4_sq *sq) { contigfree(sq->queue, sq->memsize, M_DEVBUF); } static void dealloc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq) { dealloc_host_sq(rdev, sq); } static int alloc_host_sq(struct c4iw_rdev *rdev, struct t4_sq *sq) { sq->queue = contigmalloc(sq->memsize, M_DEVBUF, M_NOWAIT, 0ul, ~0ul, 4096, 0); if (sq->queue) sq->dma_addr = vtophys(sq->queue); else return -ENOMEM; sq->phys_addr = vtophys(sq->queue); pci_unmap_addr_set(sq, mapping, sq->dma_addr); CTR4(KTR_IW_CXGBE, "%s sq %p dma_addr %p phys_addr %p", __func__, sq->queue, sq->dma_addr, sq->phys_addr); return 0; } static int destroy_qp(struct c4iw_rdev *rdev, struct t4_wq *wq, struct c4iw_dev_ucontext *uctx) { /* * uP clears EQ contexts when the connection exits rdma mode, * so no need to post a RESET WR for these EQs. */ contigfree(wq->rq.queue, wq->rq.memsize, M_DEVBUF); dealloc_sq(rdev, &wq->sq); 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; 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; int eqsize; struct wrqe *wr; 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) goto err1; if (!user) { wq->sq.sw_sq = kzalloc(wq->sq.size * sizeof *wq->sq.sw_sq, GFP_KERNEL); if (!wq->sq.sw_sq) goto err2; wq->rq.sw_rq = kzalloc(wq->rq.size * sizeof *wq->rq.sw_rq, GFP_KERNEL); if (!wq->rq.sw_rq) goto err3; } /* RQT must be a power of 2. */ wq->rq.rqt_size = roundup_pow_of_two(wq->rq.size); wq->rq.rqt_hwaddr = c4iw_rqtpool_alloc(rdev, wq->rq.rqt_size); if (!wq->rq.rqt_hwaddr) goto err4; if (alloc_host_sq(rdev, &wq->sq)) goto err5; memset(wq->sq.queue, 0, wq->sq.memsize); pci_unmap_addr_set(&wq->sq, mapping, wq->sq.dma_addr); wq->rq.queue = contigmalloc(wq->rq.memsize, M_DEVBUF, M_NOWAIT, 0ul, ~0ul, 4096, 0); if (wq->rq.queue) wq->rq.dma_addr = vtophys(wq->rq.queue); else goto err6; CTR5(KTR_IW_CXGBE, "%s sq base va 0x%p pa 0x%llx rq base va 0x%p pa 0x%llx", __func__, wq->sq.queue, (unsigned long long)vtophys(wq->sq.queue), wq->rq.queue, (unsigned long long)vtophys(wq->rq.queue)); memset(wq->rq.queue, 0, wq->rq.memsize); pci_unmap_addr_set(&wq->rq, mapping, wq->rq.dma_addr); wq->db = (void *)((unsigned long)rman_get_virtual(sc->regs_res) + MYPF_REG(SGE_PF_KDOORBELL)); wq->gts = (void *)((unsigned long)rman_get_virtual(rdev->adap->regs_res) + MYPF_REG(SGE_PF_GTS)); if (user) { wq->sq.udb = (u64)((char*)rman_get_virtual(rdev->adap->udbs_res) + (wq->sq.qid << rdev->qpshift)); wq->sq.udb &= PAGE_MASK; wq->rq.udb = (u64)((char*)rman_get_virtual(rdev->adap->udbs_res) + (wq->rq.qid << rdev->qpshift)); wq->rq.udb &= PAGE_MASK; } 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.mgmtq); if (wr == NULL) return (0); 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 + spg_creds; 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(2) | 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 + spg_creds ; 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(2) | 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, __func__); if (ret) goto err7; CTR6(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, wq->db, (unsigned long long)wq->sq.udb, (unsigned long long)wq->rq.udb); return 0; err7: contigfree(wq->rq.queue, wq->rq.memsize, M_DEVBUF); err6: dealloc_sq(rdev, &wq->sq); err5: c4iw_rqtpool_free(rdev, wq->rq.rqt_hwaddr, wq->rq.rqt_size); err4: kfree(wq->rq.sw_rq); err3: kfree(wq->sq.sw_sq); err2: c4iw_put_qpid(rdev, wq->rq.qid, uctx); err1: c4iw_put_qpid(rdev, wq->sq.qid, uctx); return -ENOMEM; } 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; } 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.r2 = 0; wqe->write.stag_sink = cpu_to_be32(wr->wr.rdma.rkey); wqe->write.to_sink = cpu_to_be64(wr->wr.rdma.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) { wqe->read.stag_src = cpu_to_be32(wr->wr.rdma.rkey); wqe->read.to_src_hi = cpu_to_be32((u32)(wr->wr.rdma.remote_addr >> 32)); wqe->read.to_src_lo = cpu_to_be32((u32)wr->wr.rdma.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_fastreg(struct t4_sq *sq, union t4_wr *wqe, struct ib_send_wr *wr, u8 *len16) { struct fw_ri_immd *imdp; __be64 *p; int i; int pbllen = roundup(wr->wr.fast_reg.page_list_len * sizeof(u64), 32); int rem; if (wr->wr.fast_reg.page_list_len > T4_MAX_FR_DEPTH) return -EINVAL; wqe->fr.qpbinde_to_dcacpu = 0; wqe->fr.pgsz_shift = wr->wr.fast_reg.page_shift - 12; wqe->fr.addr_type = FW_RI_VA_BASED_TO; wqe->fr.mem_perms = c4iw_ib_to_tpt_access(wr->wr.fast_reg.access_flags); wqe->fr.len_hi = 0; wqe->fr.len_lo = cpu_to_be32(wr->wr.fast_reg.length); wqe->fr.stag = cpu_to_be32(wr->wr.fast_reg.rkey); wqe->fr.va_hi = cpu_to_be32(wr->wr.fast_reg.iova_start >> 32); wqe->fr.va_lo_fbo = cpu_to_be32(wr->wr.fast_reg.iova_start & 0xffffffff); WARN_ON(pbllen > T4_MAX_FR_IMMD); 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 < wr->wr.fast_reg.page_list_len; i++) { *p = cpu_to_be64((u64)wr->wr.fast_reg.page_list->page_list[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; } 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; } void c4iw_qp_add_ref(struct ib_qp *qp) { CTR2(KTR_IW_CXGBE, "%s ib_qp %p", __func__, qp); atomic_inc(&(to_c4iw_qp(qp)->refcnt)); } void c4iw_qp_rem_ref(struct ib_qp *qp) { CTR2(KTR_IW_CXGBE, "%s ib_qp %p", __func__, qp); if (atomic_dec_and_test(&(to_c4iw_qp(qp)->refcnt))) wake_up(&(to_c4iw_qp(qp)->wait)); } 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; u32 num_wrs; struct t4_swsqe *swsqe; 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); return -EINVAL; } num_wrs = t4_sq_avail(&qhp->wq); if (num_wrs == 0) { spin_unlock_irqrestore(&qhp->lock, flag); 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) 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) 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_FAST_REG_MR: fw_opcode = FW_RI_FR_NSMR_WR; swsqe->opcode = FW_RI_FAST_REGISTER; err = build_fastreg(&qhp->wq.sq, wqe, wr, &len16); 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); 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); 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); } if (t4_wq_db_enabled(&qhp->wq)) t4_ring_sq_db(&qhp->wq, idx); 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; 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); return -EINVAL; } num_wrs = t4_rq_avail(&qhp->wq); if (num_wrs == 0) { spin_unlock_irqrestore(&qhp->lock, flag); 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--; } if (t4_wq_db_enabled(&qhp->wq)) t4_ring_rq_db(&qhp->wq, idx); spin_unlock_irqrestore(&qhp->lock, flag); return err; } int c4iw_bind_mw(struct ib_qp *qp, struct ib_mw *mw, struct ib_mw_bind *mw_bind) { return -ENOSYS; } 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) { 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); creds(toep, sizeof(*wqe)); 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 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); c4iw_flush_hw_cq(&rchp->cq); c4iw_count_rcqes(&rchp->cq, &qhp->wq, &count); flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count); spin_unlock(&qhp->lock); spin_unlock_irqrestore(&rchp->lock, flag); if (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); } /* locking hierarchy: cq lock first, then qp lock. */ spin_lock_irqsave(&schp->lock, flag); spin_lock(&qhp->lock); c4iw_flush_hw_cq(&schp->cq); c4iw_count_scqes(&schp->cq, &qhp->wq, &count); flushed = c4iw_flush_sq(&qhp->wq, &schp->cq, count); spin_unlock(&qhp->lock); spin_unlock_irqrestore(&schp->lock, flag); if (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 = get_chp(qhp->rhp, qhp->attr.rcq); schp = get_chp(qhp->rhp, qhp->attr.scq); if (qhp->ibqp.uobject) { t4_set_wq_in_error(&qhp->wq); 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__)); CTR4(KTR_IW_CXGBE, "%s qhp %p qid 0x%x tid %u", __func__, qhp, qhp->wq.sq.qid, 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); creds(toep, sizeof(*wqe)); t4_wrq_tx(sc, wr); ret = c4iw_wait_for_reply(rdev, &ep->com.wr_wait, ep->hwtid, qhp->wq.sq.qid, __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 void creds(struct toepcb *toep, size_t wrsize) { struct ofld_tx_sdesc *txsd; CTR3(KTR_IW_CXGBE, "%s:creB %p %u", __func__, toep , wrsize); INP_WLOCK(toep->inp); 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(toep->inp); CTR5(KTR_IW_CXGBE, "%s:creE %p %u %u %u", __func__, toep , txsd->tx_credits, toep->tx_credits, toep->txsd_pidx); } 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; CTR4(KTR_IW_CXGBE, "%s qhp %p qid 0x%x tid %u", __func__, qhp, qhp->wq.sq.qid, 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.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); creds(toep, sizeof(*wqe)); t4_wrq_tx(sc, wr); ret = c4iw_wait_for_reply(rdev, &ep->com.wr_wait, ep->hwtid, qhp->wq.sq.qid, __func__); toep->ulp_mode = ULP_MODE_RDMA; 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 > c4iw_max_read_depth) { 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; 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: //Fixme: Use atomic_read as same as Linux BUG_ON(qhp->ep->com.kref.count < 2); set_state(qhp, C4IW_QP_STATE_CLOSING); ep = qhp->ep; if (!internal) { abort = 0; disconnect = 1; c4iw_get_ep(&qhp->ep->com); } if (qhp->ibqp.uobject) t4_set_wq_in_error(&qhp->wq); ret = rdma_fini(rhp, qhp, ep); if (ret) goto err; break; case C4IW_QP_STATE_TERMINATE: set_state(qhp, C4IW_QP_STATE_TERMINATE); qhp->attr.layer_etype = attrs->layer_etype; qhp->attr.ecode = attrs->ecode; if (qhp->ibqp.uobject) t4_set_wq_in_error(&qhp->wq); ep = qhp->ep; if (!internal) terminate = 1; disconnect = 1; c4iw_get_ep(&qhp->ep->com); break; case C4IW_QP_STATE_ERROR: set_state(qhp, C4IW_QP_STATE_ERROR); if (qhp->ibqp.uobject) t4_set_wq_in_error(&qhp->wq); 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: if (!internal) { 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; - wake_up(&qhp->wait); 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; } static int enable_qp_db(int id, void *p, void *data) { struct c4iw_qp *qp = p; t4_enable_wq_db(&qp->wq); return 0; } int c4iw_destroy_qp(struct ib_qp *ib_qp) { struct c4iw_dev *rhp; struct c4iw_qp *qhp; struct c4iw_qp_attributes attrs; struct c4iw_ucontext *ucontext; 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); spin_lock_irq(&rhp->lock); remove_handle_nolock(rhp, &rhp->qpidr, qhp->wq.sq.qid); rhp->qpcnt--; BUG_ON(rhp->qpcnt < 0); if (rhp->qpcnt <= db_fc_threshold && rhp->db_state == FLOW_CONTROL) { rhp->rdev.stats.db_state_transitions++; rhp->db_state = NORMAL; idr_for_each(&rhp->qpidr, enable_qp_db, NULL); } spin_unlock_irq(&rhp->lock); atomic_dec(&qhp->refcnt); wait_event(qhp->wait, !atomic_read(&qhp->refcnt)); ucontext = ib_qp->uobject ? to_c4iw_ucontext(ib_qp->uobject->context) : NULL; destroy_qp(&rhp->rdev, &qhp->wq, ucontext ? &ucontext->uctx : &rhp->rdev.uctx); CTR3(KTR_IW_CXGBE, "%s ib_qp %p qpid 0x%0x", __func__, ib_qp, qhp->wq.sq.qid); kfree(qhp); return 0; } static int disable_qp_db(int id, void *p, void *data) { struct c4iw_qp *qp = p; t4_disable_wq_db(&qp->wq); 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; int sqsize, rqsize; struct c4iw_ucontext *ucontext; int ret; struct c4iw_mm_entry *mm1, *mm2, *mm3, *mm4; 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); rqsize = roundup(attrs->cap.max_recv_wr + 1, 16); if (rqsize > T4_MAX_RQ_SIZE) return ERR_PTR(-E2BIG); sqsize = roundup(attrs->cap.max_send_wr + 1, 16); if (sqsize > T4_MAX_SQ_SIZE) return ERR_PTR(-E2BIG); 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 + 1) * sizeof *qhp->wq.sq.queue; qhp->wq.rq.size = rqsize; qhp->wq.rq.memsize = (rqsize + 1) * 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 = 1; qhp->attr.max_ird = 1; spin_lock_init(&qhp->lock); mutex_init(&qhp->mutex); init_waitqueue_head(&qhp->wait); atomic_set(&qhp->refcnt, 1); spin_lock_irq(&rhp->lock); if (rhp->db_state != NORMAL) t4_disable_wq_db(&qhp->wq); if (++rhp->qpcnt > db_fc_threshold && rhp->db_state == NORMAL) { rhp->rdev.stats.db_state_transitions++; rhp->db_state = FLOW_CONTROL; idr_for_each(&rhp->qpidr, disable_qp_db, NULL); } ret = insert_handle_nolock(rhp, &rhp->qpidr, qhp, qhp->wq.sq.qid); spin_unlock_irq(&rhp->lock); if (ret) goto err2; if (udata) { mm1 = kmalloc(sizeof *mm1, GFP_KERNEL); if (!mm1) { ret = -ENOMEM; goto err3; } mm2 = kmalloc(sizeof *mm2, GFP_KERNEL); if (!mm2) { ret = -ENOMEM; goto err4; } mm3 = kmalloc(sizeof *mm3, GFP_KERNEL); if (!mm3) { ret = -ENOMEM; goto err5; } mm4 = kmalloc(sizeof *mm4, GFP_KERNEL); if (!mm4) { 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.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; mm1->key = uresp.sq_key; mm1->addr = qhp->wq.sq.phys_addr; mm1->len = PAGE_ALIGN(qhp->wq.sq.memsize); CTR4(KTR_IW_CXGBE, "%s mm1 %x, %x, %d", __func__, mm1->key, mm1->addr, mm1->len); insert_mmap(ucontext, mm1); mm2->key = uresp.rq_key; mm2->addr = vtophys(qhp->wq.rq.queue); mm2->len = PAGE_ALIGN(qhp->wq.rq.memsize); CTR4(KTR_IW_CXGBE, "%s mm2 %x, %x, %d", __func__, mm2->key, mm2->addr, mm2->len); insert_mmap(ucontext, mm2); mm3->key = uresp.sq_db_gts_key; mm3->addr = qhp->wq.sq.udb; mm3->len = PAGE_SIZE; CTR4(KTR_IW_CXGBE, "%s mm3 %x, %x, %d", __func__, mm3->key, mm3->addr, mm3->len); insert_mmap(ucontext, mm3); mm4->key = uresp.rq_db_gts_key; mm4->addr = qhp->wq.rq.udb; mm4->len = PAGE_SIZE; CTR4(KTR_IW_CXGBE, "%s mm4 %x, %x, %d", __func__, mm4->key, mm4->addr, mm4->len); insert_mmap(ucontext, mm4); } qhp->ibqp.qp_num = qhp->wq.sq.qid; init_timer(&(qhp->timer)); CTR5(KTR_IW_CXGBE, "%s qhp %p sq_num_entries %d, rq_num_entries %d qpid 0x%0x", __func__, qhp, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries, qhp->wq.sq.qid); return &qhp->ibqp; err7: kfree(mm4); err6: kfree(mm3); err5: kfree(mm2); err4: kfree(mm1); 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; /* * Use SQ_PSN and RQ_PSN to pass in IDX_INC values for * ringing the queue db when we're in DB_FULL mode. */ attrs.sq_db_inc = attr->sq_psn; attrs.rq_db_inc = attr->rq_psn; mask |= (attr_mask & IB_QP_SQ_PSN) ? C4IW_QP_ATTR_SQ_DB : 0; mask |= (attr_mask & IB_QP_RQ_PSN) ? C4IW_QP_ATTR_RQ_DB : 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); return 0; } #endif Index: stable/10/sys/dev/cxgbe/tom/t4_cpl_io.c =================================================================== --- stable/10/sys/dev/cxgbe/tom/t4_cpl_io.c (revision 283855) +++ stable/10/sys/dev/cxgbe/tom/t4_cpl_io.c (revision 283856) @@ -1,1834 +1,1834 @@ /*- * Copyright (c) 2012 Chelsio Communications, Inc. * All rights reserved. * Written by: Navdeep Parhar * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. 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. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #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 #define TCPSTATES #include #include #include #include "common/common.h" #include "common/t4_msg.h" #include "common/t4_regs.h" #include "common/t4_tcb.h" #include "tom/t4_tom_l2t.h" #include "tom/t4_tom.h" VNET_DECLARE(int, tcp_do_autosndbuf); #define V_tcp_do_autosndbuf VNET(tcp_do_autosndbuf) VNET_DECLARE(int, tcp_autosndbuf_inc); #define V_tcp_autosndbuf_inc VNET(tcp_autosndbuf_inc) VNET_DECLARE(int, tcp_autosndbuf_max); #define V_tcp_autosndbuf_max VNET(tcp_autosndbuf_max) VNET_DECLARE(int, tcp_do_autorcvbuf); #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) VNET_DECLARE(int, tcp_autorcvbuf_inc); #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) VNET_DECLARE(int, tcp_autorcvbuf_max); #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) /* * For ULP connections HW may add headers, e.g., for digests, that aren't part * of the messages sent by the host but that are part of the TCP payload and * therefore consume TCP sequence space. Tx connection parameters that * operate in TCP sequence space are affected by the HW additions and need to * compensate for them to accurately track TCP sequence numbers. This array * contains the compensating extra lengths for ULP packets. It is indexed by * a packet's ULP submode. */ const unsigned int t4_ulp_extra_len[] = {0, 4, 4, 8}; /* * Return the length of any HW additions that will be made to a Tx packet. * Such additions can happen for some types of ULP packets. */ static inline unsigned int ulp_extra_len(struct mbuf *m, int *ulp_mode) { struct m_tag *mtag; if ((mtag = m_tag_find(m, CXGBE_ISCSI_MBUF_TAG, NULL)) == NULL) return (0); *ulp_mode = *((int *)(mtag + 1)); return (t4_ulp_extra_len[*ulp_mode & 3]); } void send_flowc_wr(struct toepcb *toep, struct flowc_tx_params *ftxp) { struct wrqe *wr; struct fw_flowc_wr *flowc; unsigned int nparams = ftxp ? 8 : 6, flowclen; struct port_info *pi = toep->port; struct adapter *sc = pi->adapter; unsigned int pfvf = G_FW_VIID_PFN(pi->viid) << S_FW_VIID_PFN; struct ofld_tx_sdesc *txsd = &toep->txsd[toep->txsd_pidx]; KASSERT(!(toep->flags & TPF_FLOWC_WR_SENT), ("%s: flowc for tid %u sent already", __func__, toep->tid)); flowclen = sizeof(*flowc) + nparams * sizeof(struct fw_flowc_mnemval); wr = alloc_wrqe(roundup2(flowclen, 16), toep->ofld_txq); if (wr == NULL) { /* XXX */ panic("%s: allocation failure.", __func__); } flowc = wrtod(wr); memset(flowc, 0, wr->wr_len); flowc->op_to_nparams = htobe32(V_FW_WR_OP(FW_FLOWC_WR) | V_FW_FLOWC_WR_NPARAMS(nparams)); flowc->flowid_len16 = htonl(V_FW_WR_LEN16(howmany(flowclen, 16)) | V_FW_WR_FLOWID(toep->tid)); flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN; flowc->mnemval[0].val = htobe32(pfvf); flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH; flowc->mnemval[1].val = htobe32(pi->tx_chan); flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT; flowc->mnemval[2].val = htobe32(pi->tx_chan); flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID; flowc->mnemval[3].val = htobe32(toep->ofld_rxq->iq.abs_id); if (ftxp) { uint32_t sndbuf = min(ftxp->snd_space, sc->tt.sndbuf); flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT; flowc->mnemval[4].val = htobe32(ftxp->snd_nxt); flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT; flowc->mnemval[5].val = htobe32(ftxp->rcv_nxt); flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF; flowc->mnemval[6].val = htobe32(sndbuf); flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS; flowc->mnemval[7].val = htobe32(ftxp->mss); CTR6(KTR_CXGBE, "%s: tid %u, mss %u, sndbuf %u, snd_nxt 0x%x, rcv_nxt 0x%x", __func__, toep->tid, ftxp->mss, sndbuf, ftxp->snd_nxt, ftxp->rcv_nxt); } else { flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDBUF; flowc->mnemval[4].val = htobe32(512); flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_MSS; flowc->mnemval[5].val = htobe32(512); CTR2(KTR_CXGBE, "%s: tid %u", __func__, toep->tid); } txsd->tx_credits = howmany(flowclen, 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--; toep->flags |= TPF_FLOWC_WR_SENT; t4_wrq_tx(sc, wr); } void send_reset(struct adapter *sc, struct toepcb *toep, uint32_t snd_nxt) { struct wrqe *wr; struct cpl_abort_req *req; int tid = toep->tid; struct inpcb *inp = toep->inp; struct tcpcb *tp = intotcpcb(inp); /* don't use if INP_DROPPED */ INP_WLOCK_ASSERT(inp); CTR6(KTR_CXGBE, "%s: tid %d (%s), toep_flags 0x%x, inp_flags 0x%x%s", __func__, toep->tid, inp->inp_flags & INP_DROPPED ? "inp dropped" : tcpstates[tp->t_state], toep->flags, inp->inp_flags, toep->flags & TPF_ABORT_SHUTDOWN ? " (abort already in progress)" : ""); if (toep->flags & TPF_ABORT_SHUTDOWN) return; /* abort already in progress */ toep->flags |= TPF_ABORT_SHUTDOWN; KASSERT(toep->flags & TPF_FLOWC_WR_SENT, ("%s: flowc_wr not sent for tid %d.", __func__, tid)); wr = alloc_wrqe(sizeof(*req), toep->ofld_txq); if (wr == NULL) { /* XXX */ panic("%s: allocation failure.", __func__); } req = wrtod(wr); INIT_TP_WR_MIT_CPL(req, CPL_ABORT_REQ, tid); if (inp->inp_flags & INP_DROPPED) req->rsvd0 = htobe32(snd_nxt); else req->rsvd0 = htobe32(tp->snd_nxt); req->rsvd1 = !(toep->flags & TPF_TX_DATA_SENT); req->cmd = CPL_ABORT_SEND_RST; /* * XXX: What's the correct way to tell that the inp hasn't been detached * from its socket? Should I even be flushing the snd buffer here? */ if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) == 0) { struct socket *so = inp->inp_socket; if (so != NULL) /* because I'm not sure. See comment above */ sbflush(&so->so_snd); } t4_l2t_send(sc, wr, toep->l2te); } /* * Called when a connection is established to translate the TCP options * reported by HW to FreeBSD's native format. */ static void assign_rxopt(struct tcpcb *tp, unsigned int opt) { struct toepcb *toep = tp->t_toe; struct inpcb *inp = tp->t_inpcb; struct adapter *sc = td_adapter(toep->td); int n; INP_LOCK_ASSERT(inp); if (inp->inp_inc.inc_flags & INC_ISIPV6) n = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); else n = sizeof(struct ip) + sizeof(struct tcphdr); tp->t_maxseg = tp->t_maxopd = sc->params.mtus[G_TCPOPT_MSS(opt)] - n; CTR4(KTR_CXGBE, "%s: tid %d, mtu_idx %u (%u)", __func__, toep->tid, G_TCPOPT_MSS(opt), sc->params.mtus[G_TCPOPT_MSS(opt)]); if (G_TCPOPT_TSTAMP(opt)) { tp->t_flags |= TF_RCVD_TSTMP; /* timestamps ok */ tp->ts_recent = 0; /* hmmm */ tp->ts_recent_age = tcp_ts_getticks(); tp->t_maxseg -= TCPOLEN_TSTAMP_APPA; } if (G_TCPOPT_SACK(opt)) tp->t_flags |= TF_SACK_PERMIT; /* should already be set */ else tp->t_flags &= ~TF_SACK_PERMIT; /* sack disallowed by peer */ if (G_TCPOPT_WSCALE_OK(opt)) tp->t_flags |= TF_RCVD_SCALE; /* Doing window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == (TF_RCVD_SCALE | TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; tp->snd_scale = G_TCPOPT_SND_WSCALE(opt); } } /* * Completes some final bits of initialization for just established connections * and changes their state to TCPS_ESTABLISHED. * * The ISNs are from after the exchange of SYNs. i.e., the true ISN + 1. */ void make_established(struct toepcb *toep, uint32_t snd_isn, uint32_t rcv_isn, uint16_t opt) { struct inpcb *inp = toep->inp; struct socket *so = inp->inp_socket; struct tcpcb *tp = intotcpcb(inp); long bufsize; uint32_t iss = be32toh(snd_isn) - 1; /* true ISS */ uint32_t irs = be32toh(rcv_isn) - 1; /* true IRS */ uint16_t tcpopt = be16toh(opt); struct flowc_tx_params ftxp; INP_WLOCK_ASSERT(inp); KASSERT(tp->t_state == TCPS_SYN_SENT || tp->t_state == TCPS_SYN_RECEIVED, ("%s: TCP state %s", __func__, tcpstates[tp->t_state])); CTR4(KTR_CXGBE, "%s: tid %d, toep %p, inp %p", __func__, toep->tid, toep, inp); tp->t_state = TCPS_ESTABLISHED; tp->t_starttime = ticks; TCPSTAT_INC(tcps_connects); tp->irs = irs; tcp_rcvseqinit(tp); tp->rcv_wnd = toep->rx_credits << 10; tp->rcv_adv += tp->rcv_wnd; tp->last_ack_sent = tp->rcv_nxt; /* * If we were unable to send all rx credits via opt0, save the remainder * in rx_credits so that they can be handed over with the next credit * update. */ SOCKBUF_LOCK(&so->so_rcv); bufsize = select_rcv_wnd(so); SOCKBUF_UNLOCK(&so->so_rcv); toep->rx_credits = bufsize - tp->rcv_wnd; tp->iss = iss; tcp_sendseqinit(tp); tp->snd_una = iss + 1; tp->snd_nxt = iss + 1; tp->snd_max = iss + 1; assign_rxopt(tp, tcpopt); SOCKBUF_LOCK(&so->so_snd); if (so->so_snd.sb_flags & SB_AUTOSIZE && V_tcp_do_autosndbuf) bufsize = V_tcp_autosndbuf_max; else bufsize = sbspace(&so->so_snd); SOCKBUF_UNLOCK(&so->so_snd); ftxp.snd_nxt = tp->snd_nxt; ftxp.rcv_nxt = tp->rcv_nxt; ftxp.snd_space = bufsize; ftxp.mss = tp->t_maxseg; send_flowc_wr(toep, &ftxp); soisconnected(so); } static int send_rx_credits(struct adapter *sc, struct toepcb *toep, int credits) { struct wrqe *wr; struct cpl_rx_data_ack *req; uint32_t dack = F_RX_DACK_CHANGE | V_RX_DACK_MODE(1); KASSERT(credits >= 0, ("%s: %d credits", __func__, credits)); wr = alloc_wrqe(sizeof(*req), toep->ctrlq); if (wr == NULL) return (0); req = wrtod(wr); INIT_TP_WR_MIT_CPL(req, CPL_RX_DATA_ACK, toep->tid); req->credit_dack = htobe32(dack | V_RX_CREDITS(credits)); t4_wrq_tx(sc, wr); return (credits); } void t4_rcvd(struct toedev *tod, struct tcpcb *tp) { struct adapter *sc = tod->tod_softc; struct inpcb *inp = tp->t_inpcb; struct socket *so = inp->inp_socket; struct sockbuf *sb = &so->so_rcv; struct toepcb *toep = tp->t_toe; int credits; INP_WLOCK_ASSERT(inp); SOCKBUF_LOCK(sb); KASSERT(toep->sb_cc >= sb->sb_cc, ("%s: sb %p has more data (%d) than last time (%d).", __func__, sb, sb->sb_cc, toep->sb_cc)); if (toep->ulp_mode == ULP_MODE_ISCSI) { toep->rx_credits += toep->sb_cc; toep->sb_cc = 0; } else { toep->rx_credits += toep->sb_cc - sb->sb_cc; toep->sb_cc = sb->sb_cc; } credits = toep->rx_credits; SOCKBUF_UNLOCK(sb); if (credits > 0 && (credits + 16384 >= tp->rcv_wnd || credits >= 15 * 1024)) { credits = send_rx_credits(sc, toep, credits); SOCKBUF_LOCK(sb); toep->rx_credits -= credits; SOCKBUF_UNLOCK(sb); tp->rcv_wnd += credits; tp->rcv_adv += credits; } } /* * Close a connection by sending a CPL_CLOSE_CON_REQ message. */ static int close_conn(struct adapter *sc, struct toepcb *toep) { struct wrqe *wr; struct cpl_close_con_req *req; unsigned int tid = toep->tid; CTR3(KTR_CXGBE, "%s: tid %u%s", __func__, toep->tid, toep->flags & TPF_FIN_SENT ? ", IGNORED" : ""); if (toep->flags & TPF_FIN_SENT) return (0); KASSERT(toep->flags & TPF_FLOWC_WR_SENT, ("%s: flowc_wr not sent for tid %u.", __func__, tid)); wr = alloc_wrqe(sizeof(*req), toep->ofld_txq); if (wr == NULL) { /* XXX */ panic("%s: allocation failure.", __func__); } req = wrtod(wr); req->wr.wr_hi = htonl(V_FW_WR_OP(FW_TP_WR) | V_FW_WR_IMMDLEN(sizeof(*req) - sizeof(req->wr))); req->wr.wr_mid = htonl(V_FW_WR_LEN16(howmany(sizeof(*req), 16)) | V_FW_WR_FLOWID(tid)); req->wr.wr_lo = cpu_to_be64(0); OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid)); req->rsvd = 0; toep->flags |= TPF_FIN_SENT; toep->flags &= ~TPF_SEND_FIN; t4_l2t_send(sc, wr, toep->l2te); return (0); } #define MAX_OFLD_TX_CREDITS (SGE_MAX_WR_LEN / 16) #define MIN_OFLD_TX_CREDITS (howmany(sizeof(struct fw_ofld_tx_data_wr) + 1, 16)) /* Maximum amount of immediate data we could stuff in a WR */ static inline int max_imm_payload(int tx_credits) { const int n = 2; /* Use only up to 2 desc for imm. data WR */ KASSERT(tx_credits >= 0 && tx_credits <= MAX_OFLD_TX_CREDITS, ("%s: %d credits", __func__, tx_credits)); if (tx_credits < MIN_OFLD_TX_CREDITS) return (0); if (tx_credits >= (n * EQ_ESIZE) / 16) return ((n * EQ_ESIZE) - sizeof(struct fw_ofld_tx_data_wr)); else return (tx_credits * 16 - sizeof(struct fw_ofld_tx_data_wr)); } /* Maximum number of SGL entries we could stuff in a WR */ static inline int max_dsgl_nsegs(int tx_credits) { int nseg = 1; /* ulptx_sgl has room for 1, rest ulp_tx_sge_pair */ int sge_pair_credits = tx_credits - MIN_OFLD_TX_CREDITS; KASSERT(tx_credits >= 0 && tx_credits <= MAX_OFLD_TX_CREDITS, ("%s: %d credits", __func__, tx_credits)); if (tx_credits < MIN_OFLD_TX_CREDITS) return (0); nseg += 2 * (sge_pair_credits * 16 / 24); if ((sge_pair_credits * 16) % 24 == 16) nseg++; return (nseg); } static inline void write_tx_wr(void *dst, struct toepcb *toep, unsigned int immdlen, unsigned int plen, uint8_t credits, int shove, int ulp_mode, int txalign) { struct fw_ofld_tx_data_wr *txwr = dst; unsigned int wr_ulp_mode; txwr->op_to_immdlen = htobe32(V_WR_OP(FW_OFLD_TX_DATA_WR) | V_FW_WR_IMMDLEN(immdlen)); txwr->flowid_len16 = htobe32(V_FW_WR_FLOWID(toep->tid) | V_FW_WR_LEN16(credits)); /* for iscsi, the mode & submode setting is per-packet */ if (toep->ulp_mode == ULP_MODE_ISCSI) wr_ulp_mode = V_FW_OFLD_TX_DATA_WR_ULPMODE(ulp_mode >> 4) | V_FW_OFLD_TX_DATA_WR_ULPSUBMODE(ulp_mode & 3); else wr_ulp_mode = V_FW_OFLD_TX_DATA_WR_ULPMODE(toep->ulp_mode); txwr->lsodisable_to_proxy = htobe32(wr_ulp_mode | V_FW_OFLD_TX_DATA_WR_URGENT(0) | /* XXX */ V_FW_OFLD_TX_DATA_WR_SHOVE(shove)); txwr->plen = htobe32(plen); if (txalign > 0) { struct tcpcb *tp = intotcpcb(toep->inp); if (plen < 2 * tp->t_maxseg || is_10G_port(toep->port)) txwr->lsodisable_to_proxy |= htobe32(F_FW_OFLD_TX_DATA_WR_LSODISABLE); else txwr->lsodisable_to_proxy |= htobe32(F_FW_OFLD_TX_DATA_WR_ALIGNPLD | (tp->t_flags & TF_NODELAY ? 0 : F_FW_OFLD_TX_DATA_WR_ALIGNPLDSHOVE)); } } /* * Generate a DSGL from a starting mbuf. The total number of segments and the * maximum segments in any one mbuf are provided. */ static void write_tx_sgl(void *dst, struct mbuf *start, struct mbuf *stop, int nsegs, int n) { struct mbuf *m; struct ulptx_sgl *usgl = dst; int i, j, rc; struct sglist sg; struct sglist_seg segs[n]; KASSERT(nsegs > 0, ("%s: nsegs 0", __func__)); sglist_init(&sg, n, segs); usgl->cmd_nsge = htobe32(V_ULPTX_CMD(ULP_TX_SC_DSGL) | V_ULPTX_NSGE(nsegs)); i = -1; for (m = start; m != stop; m = m->m_next) { rc = sglist_append(&sg, mtod(m, void *), m->m_len); if (__predict_false(rc != 0)) panic("%s: sglist_append %d", __func__, rc); for (j = 0; j < sg.sg_nseg; i++, j++) { if (i < 0) { usgl->len0 = htobe32(segs[j].ss_len); usgl->addr0 = htobe64(segs[j].ss_paddr); } else { usgl->sge[i / 2].len[i & 1] = htobe32(segs[j].ss_len); usgl->sge[i / 2].addr[i & 1] = htobe64(segs[j].ss_paddr); } #ifdef INVARIANTS nsegs--; #endif } sglist_reset(&sg); } if (i & 1) usgl->sge[i / 2].len[1] = htobe32(0); KASSERT(nsegs == 0, ("%s: nsegs %d, start %p, stop %p", __func__, nsegs, start, stop)); } /* * Max number of SGL entries an offload tx work request can have. This is 41 * (1 + 40) for a full 512B work request. * fw_ofld_tx_data_wr(16B) + ulptx_sgl(16B, 1) + ulptx_sge_pair(480B, 40) */ #define OFLD_SGL_LEN (41) /* * Send data and/or a FIN to the peer. * * The socket's so_snd buffer consists of a stream of data starting with sb_mb * and linked together with m_next. sb_sndptr, if set, is the last mbuf that * was transmitted. * * drop indicates the number of bytes that should be dropped from the head of * the send buffer. It is an optimization that lets do_fw4_ack avoid creating * contention on the send buffer lock (before this change it used to do * sowwakeup and then t4_push_frames right after that when recovering from tx * stalls). When drop is set this function MUST drop the bytes and wake up any * writers. */ void t4_push_frames(struct adapter *sc, struct toepcb *toep, int drop) { struct mbuf *sndptr, *m, *sb_sndptr; struct fw_ofld_tx_data_wr *txwr; struct wrqe *wr; u_int plen, nsegs, credits, max_imm, max_nsegs, max_nsegs_1mbuf; struct inpcb *inp = toep->inp; struct tcpcb *tp = intotcpcb(inp); struct socket *so = inp->inp_socket; struct sockbuf *sb = &so->so_snd; int tx_credits, shove, compl, space, sowwakeup; struct ofld_tx_sdesc *txsd = &toep->txsd[toep->txsd_pidx]; INP_WLOCK_ASSERT(inp); KASSERT(toep->flags & TPF_FLOWC_WR_SENT, ("%s: flowc_wr not sent for tid %u.", __func__, toep->tid)); KASSERT(toep->ulp_mode == ULP_MODE_NONE || toep->ulp_mode == ULP_MODE_TCPDDP || toep->ulp_mode == ULP_MODE_RDMA, ("%s: ulp_mode %u for toep %p", __func__, toep->ulp_mode, toep)); /* * This function doesn't resume by itself. Someone else must clear the * flag and call this function. */ if (__predict_false(toep->flags & TPF_TX_SUSPENDED)) { KASSERT(drop == 0, ("%s: drop (%d) != 0 but tx is suspended", __func__, drop)); return; } do { tx_credits = min(toep->tx_credits, MAX_OFLD_TX_CREDITS); max_imm = max_imm_payload(tx_credits); max_nsegs = max_dsgl_nsegs(tx_credits); SOCKBUF_LOCK(sb); sowwakeup = drop; if (drop) { sbdrop_locked(sb, drop); drop = 0; } sb_sndptr = sb->sb_sndptr; sndptr = sb_sndptr ? sb_sndptr->m_next : sb->sb_mb; plen = 0; nsegs = 0; max_nsegs_1mbuf = 0; /* max # of SGL segments in any one mbuf */ for (m = sndptr; m != NULL; m = m->m_next) { int n = sglist_count(mtod(m, void *), m->m_len); nsegs += n; plen += m->m_len; /* This mbuf sent us _over_ the nsegs limit, back out */ if (plen > max_imm && nsegs > max_nsegs) { nsegs -= n; plen -= m->m_len; if (plen == 0) { /* Too few credits */ toep->flags |= TPF_TX_SUSPENDED; if (sowwakeup) sowwakeup_locked(so); else SOCKBUF_UNLOCK(sb); SOCKBUF_UNLOCK_ASSERT(sb); return; } break; } if (max_nsegs_1mbuf < n) max_nsegs_1mbuf = n; sb_sndptr = m; /* new sb->sb_sndptr if all goes well */ /* This mbuf put us right at the max_nsegs limit */ if (plen > max_imm && nsegs == max_nsegs) { m = m->m_next; break; } } shove = m == NULL && !(tp->t_flags & TF_MORETOCOME); space = sbspace(sb); if (space <= sb->sb_hiwat * 3 / 8 && toep->plen_nocompl + plen >= sb->sb_hiwat / 4) compl = 1; else compl = 0; if (sb->sb_flags & SB_AUTOSIZE && V_tcp_do_autosndbuf && sb->sb_hiwat < V_tcp_autosndbuf_max && space < sb->sb_hiwat / 8) { int newsize = min(sb->sb_hiwat + V_tcp_autosndbuf_inc, V_tcp_autosndbuf_max); if (!sbreserve_locked(sb, newsize, so, NULL)) sb->sb_flags &= ~SB_AUTOSIZE; else sowwakeup = 1; /* room available */ } if (sowwakeup) sowwakeup_locked(so); else SOCKBUF_UNLOCK(sb); SOCKBUF_UNLOCK_ASSERT(sb); /* nothing to send */ if (plen == 0) { KASSERT(m == NULL, ("%s: nothing to send, but m != NULL", __func__)); break; } if (__predict_false(toep->flags & TPF_FIN_SENT)) panic("%s: excess tx.", __func__); if (plen <= max_imm) { /* Immediate data tx */ wr = alloc_wrqe(roundup2(sizeof(*txwr) + plen, 16), toep->ofld_txq); if (wr == NULL) { /* XXX: how will we recover from this? */ toep->flags |= TPF_TX_SUSPENDED; return; } txwr = wrtod(wr); credits = howmany(wr->wr_len, 16); write_tx_wr(txwr, toep, plen, plen, credits, shove, 0, sc->tt.tx_align); m_copydata(sndptr, 0, plen, (void *)(txwr + 1)); nsegs = 0; } else { int wr_len; /* DSGL tx */ wr_len = sizeof(*txwr) + sizeof(struct ulptx_sgl) + ((3 * (nsegs - 1)) / 2 + ((nsegs - 1) & 1)) * 8; wr = alloc_wrqe(roundup2(wr_len, 16), toep->ofld_txq); if (wr == NULL) { /* XXX: how will we recover from this? */ toep->flags |= TPF_TX_SUSPENDED; return; } txwr = wrtod(wr); credits = howmany(wr_len, 16); write_tx_wr(txwr, toep, 0, plen, credits, shove, 0, sc->tt.tx_align); write_tx_sgl(txwr + 1, sndptr, m, nsegs, max_nsegs_1mbuf); if (wr_len & 0xf) { uint64_t *pad = (uint64_t *) ((uintptr_t)txwr + wr_len); *pad = 0; } } KASSERT(toep->tx_credits >= credits, ("%s: not enough credits", __func__)); toep->tx_credits -= credits; toep->tx_nocompl += credits; toep->plen_nocompl += plen; if (toep->tx_credits <= toep->tx_total * 3 / 8 && toep->tx_nocompl >= toep->tx_total / 4) compl = 1; - if (compl) { + if (compl || toep->ulp_mode == ULP_MODE_RDMA) { txwr->op_to_immdlen |= htobe32(F_FW_WR_COMPL); toep->tx_nocompl = 0; toep->plen_nocompl = 0; } tp->snd_nxt += plen; tp->snd_max += plen; SOCKBUF_LOCK(sb); KASSERT(sb_sndptr, ("%s: sb_sndptr is NULL", __func__)); sb->sb_sndptr = sb_sndptr; SOCKBUF_UNLOCK(sb); toep->flags |= TPF_TX_DATA_SENT; if (toep->tx_credits < MIN_OFLD_TX_CREDITS) toep->flags |= TPF_TX_SUSPENDED; KASSERT(toep->txsd_avail > 0, ("%s: no txsd", __func__)); txsd->plen = plen; txsd->tx_credits = credits; txsd++; if (__predict_false(++toep->txsd_pidx == toep->txsd_total)) { toep->txsd_pidx = 0; txsd = &toep->txsd[0]; } toep->txsd_avail--; t4_l2t_send(sc, wr, toep->l2te); } while (m != NULL); /* Send a FIN if requested, but only if there's no more data to send */ if (m == NULL && toep->flags & TPF_SEND_FIN) close_conn(sc, toep); } /* Send ULP data over TOE using TX_DATA_WR. We send whole mbuf at once */ void t4_ulp_push_frames(struct adapter *sc, struct toepcb *toep, int drop) { struct mbuf *sndptr, *m = NULL; struct fw_ofld_tx_data_wr *txwr; struct wrqe *wr; unsigned int plen, nsegs, credits, max_imm, max_nsegs, max_nsegs_1mbuf; struct inpcb *inp = toep->inp; struct tcpcb *tp; struct socket *so; struct sockbuf *sb; int tx_credits, ulp_len = 0, ulp_mode = 0, qlen = 0; int shove, compl; struct ofld_tx_sdesc *txsd; INP_WLOCK_ASSERT(inp); if (toep->flags & TPF_ABORT_SHUTDOWN) return; tp = intotcpcb(inp); so = inp->inp_socket; sb = &so->so_snd; txsd = &toep->txsd[toep->txsd_pidx]; KASSERT(toep->flags & TPF_FLOWC_WR_SENT, ("%s: flowc_wr not sent for tid %u.", __func__, toep->tid)); /* * This function doesn't resume by itself. Someone else must clear the * flag and call this function. */ if (__predict_false(toep->flags & TPF_TX_SUSPENDED)) return; sndptr = t4_queue_iscsi_callback(so, toep, 1, &qlen); if (!qlen) return; do { tx_credits = min(toep->tx_credits, MAX_OFLD_TX_CREDITS); max_imm = max_imm_payload(tx_credits); max_nsegs = max_dsgl_nsegs(tx_credits); if (drop) { t4_cpl_iscsi_callback(toep->td, toep, &drop, CPL_FW4_ACK); drop = 0; } plen = 0; nsegs = 0; max_nsegs_1mbuf = 0; /* max # of SGL segments in any one mbuf */ for (m = sndptr; m != NULL; m = m->m_next) { int n = sglist_count(mtod(m, void *), m->m_len); nsegs += n; plen += m->m_len; /* This mbuf sent us _over_ the nsegs limit, return */ if (plen > max_imm && nsegs > max_nsegs) { toep->flags |= TPF_TX_SUSPENDED; return; } if (max_nsegs_1mbuf < n) max_nsegs_1mbuf = n; /* This mbuf put us right at the max_nsegs limit */ if (plen > max_imm && nsegs == max_nsegs) { toep->flags |= TPF_TX_SUSPENDED; return; } } shove = m == NULL && !(tp->t_flags & TF_MORETOCOME); /* nothing to send */ if (plen == 0) { KASSERT(m == NULL, ("%s: nothing to send, but m != NULL", __func__)); break; } if (__predict_false(toep->flags & TPF_FIN_SENT)) panic("%s: excess tx.", __func__); ulp_len = plen + ulp_extra_len(sndptr, &ulp_mode); if (plen <= max_imm) { /* Immediate data tx */ wr = alloc_wrqe(roundup(sizeof(*txwr) + plen, 16), toep->ofld_txq); if (wr == NULL) { /* XXX: how will we recover from this? */ toep->flags |= TPF_TX_SUSPENDED; return; } txwr = wrtod(wr); credits = howmany(wr->wr_len, 16); write_tx_wr(txwr, toep, plen, ulp_len, credits, shove, ulp_mode, 0); m_copydata(sndptr, 0, plen, (void *)(txwr + 1)); } else { int wr_len; /* DSGL tx */ wr_len = sizeof(*txwr) + sizeof(struct ulptx_sgl) + ((3 * (nsegs - 1)) / 2 + ((nsegs - 1) & 1)) * 8; wr = alloc_wrqe(roundup(wr_len, 16), toep->ofld_txq); if (wr == NULL) { /* XXX: how will we recover from this? */ toep->flags |= TPF_TX_SUSPENDED; return; } txwr = wrtod(wr); credits = howmany(wr_len, 16); write_tx_wr(txwr, toep, 0, ulp_len, credits, shove, ulp_mode, 0); write_tx_sgl(txwr + 1, sndptr, m, nsegs, max_nsegs_1mbuf); if (wr_len & 0xf) { uint64_t *pad = (uint64_t *) ((uintptr_t)txwr + wr_len); *pad = 0; } } KASSERT(toep->tx_credits >= credits, ("%s: not enough credits", __func__)); toep->tx_credits -= credits; toep->tx_nocompl += credits; toep->plen_nocompl += plen; if (toep->tx_credits <= toep->tx_total * 3 / 8 && toep->tx_nocompl >= toep->tx_total / 4) compl = 1; if (compl) { txwr->op_to_immdlen |= htobe32(F_FW_WR_COMPL); toep->tx_nocompl = 0; toep->plen_nocompl = 0; } tp->snd_nxt += ulp_len; tp->snd_max += ulp_len; /* goto next mbuf */ sndptr = m = t4_queue_iscsi_callback(so, toep, 2, &qlen); toep->flags |= TPF_TX_DATA_SENT; if (toep->tx_credits < MIN_OFLD_TX_CREDITS) { toep->flags |= TPF_TX_SUSPENDED; } KASSERT(toep->txsd_avail > 0, ("%s: no txsd", __func__)); txsd->plen = plen; txsd->tx_credits = credits; txsd++; if (__predict_false(++toep->txsd_pidx == toep->txsd_total)) { toep->txsd_pidx = 0; txsd = &toep->txsd[0]; } toep->txsd_avail--; t4_l2t_send(sc, wr, toep->l2te); } while (m != NULL); /* Send a FIN if requested, but only if there's no more data to send */ if (m == NULL && toep->flags & TPF_SEND_FIN) close_conn(sc, toep); } int t4_tod_output(struct toedev *tod, struct tcpcb *tp) { struct adapter *sc = tod->tod_softc; #ifdef INVARIANTS struct inpcb *inp = tp->t_inpcb; #endif struct toepcb *toep = tp->t_toe; INP_WLOCK_ASSERT(inp); KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("%s: inp %p dropped.", __func__, inp)); KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); t4_push_frames(sc, toep, 0); return (0); } int t4_send_fin(struct toedev *tod, struct tcpcb *tp) { struct adapter *sc = tod->tod_softc; #ifdef INVARIANTS struct inpcb *inp = tp->t_inpcb; #endif struct toepcb *toep = tp->t_toe; INP_WLOCK_ASSERT(inp); KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("%s: inp %p dropped.", __func__, inp)); KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); toep->flags |= TPF_SEND_FIN; if (tp->t_state >= TCPS_ESTABLISHED) { if (toep->ulp_mode == ULP_MODE_ISCSI) t4_ulp_push_frames(sc, toep, 0); else t4_push_frames(sc, toep, 0); } return (0); } int t4_send_rst(struct toedev *tod, struct tcpcb *tp) { struct adapter *sc = tod->tod_softc; #if defined(INVARIANTS) struct inpcb *inp = tp->t_inpcb; #endif struct toepcb *toep = tp->t_toe; INP_WLOCK_ASSERT(inp); KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("%s: inp %p dropped.", __func__, inp)); KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); /* hmmmm */ KASSERT(toep->flags & TPF_FLOWC_WR_SENT, ("%s: flowc for tid %u [%s] not sent already", __func__, toep->tid, tcpstates[tp->t_state])); send_reset(sc, toep, 0); return (0); } /* * Peer has sent us a FIN. */ static int do_peer_close(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_peer_close *cpl = (const void *)(rss + 1); unsigned int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct inpcb *inp = toep->inp; struct tcpcb *tp = NULL; struct socket *so; struct sockbuf *sb; #ifdef INVARIANTS unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); #endif KASSERT(opcode == CPL_PEER_CLOSE, ("%s: unexpected opcode 0x%x", __func__, opcode)); KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); if (__predict_false(toep->flags & TPF_SYNQE)) { #ifdef INVARIANTS struct synq_entry *synqe = (void *)toep; INP_WLOCK(synqe->lctx->inp); if (synqe->flags & TPF_SYNQE_HAS_L2TE) { KASSERT(synqe->flags & TPF_ABORT_SHUTDOWN, ("%s: listen socket closed but tid %u not aborted.", __func__, tid)); } else { /* * do_pass_accept_req is still running and will * eventually take care of this tid. */ } INP_WUNLOCK(synqe->lctx->inp); #endif CTR4(KTR_CXGBE, "%s: tid %u, synqe %p (0x%x)", __func__, tid, toep, toep->flags); return (0); } KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); INP_INFO_WLOCK(&V_tcbinfo); INP_WLOCK(inp); tp = intotcpcb(inp); CTR5(KTR_CXGBE, "%s: tid %u (%s), toep_flags 0x%x, inp %p", __func__, tid, tp ? tcpstates[tp->t_state] : "no tp", toep->flags, inp); if (toep->flags & TPF_ABORT_SHUTDOWN) goto done; tp->rcv_nxt++; /* FIN */ so = inp->inp_socket; sb = &so->so_rcv; SOCKBUF_LOCK(sb); if (__predict_false(toep->ddp_flags & (DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE))) { m = get_ddp_mbuf(be32toh(cpl->rcv_nxt) - tp->rcv_nxt); tp->rcv_nxt = be32toh(cpl->rcv_nxt); toep->ddp_flags &= ~(DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE); KASSERT(toep->sb_cc >= sb->sb_cc, ("%s: sb %p has more data (%d) than last time (%d).", __func__, sb, sb->sb_cc, toep->sb_cc)); toep->rx_credits += toep->sb_cc - sb->sb_cc; #ifdef USE_DDP_RX_FLOW_CONTROL toep->rx_credits -= m->m_len; /* adjust for F_RX_FC_DDP */ #endif sbappendstream_locked(sb, m); toep->sb_cc = sb->sb_cc; } socantrcvmore_locked(so); /* unlocks the sockbuf */ if (toep->ulp_mode != ULP_MODE_RDMA) { KASSERT(tp->rcv_nxt == be32toh(cpl->rcv_nxt), ("%s: rcv_nxt mismatch: %u %u", __func__, tp->rcv_nxt, be32toh(cpl->rcv_nxt))); } switch (tp->t_state) { case TCPS_SYN_RECEIVED: tp->t_starttime = ticks; /* FALLTHROUGH */ case TCPS_ESTABLISHED: tp->t_state = TCPS_CLOSE_WAIT; break; case TCPS_FIN_WAIT_1: tp->t_state = TCPS_CLOSING; break; case TCPS_FIN_WAIT_2: tcp_twstart(tp); INP_UNLOCK_ASSERT(inp); /* safe, we have a ref on the inp */ INP_INFO_WUNLOCK(&V_tcbinfo); INP_WLOCK(inp); final_cpl_received(toep); return (0); default: log(LOG_ERR, "%s: TID %u received CPL_PEER_CLOSE in state %d\n", __func__, tid, tp->t_state); } done: INP_WUNLOCK(inp); INP_INFO_WUNLOCK(&V_tcbinfo); return (0); } /* * Peer has ACK'd our FIN. */ static int do_close_con_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_close_con_rpl *cpl = (const void *)(rss + 1); unsigned int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct inpcb *inp = toep->inp; struct tcpcb *tp = NULL; struct socket *so = NULL; #ifdef INVARIANTS unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); #endif KASSERT(opcode == CPL_CLOSE_CON_RPL, ("%s: unexpected opcode 0x%x", __func__, opcode)); KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); INP_INFO_WLOCK(&V_tcbinfo); INP_WLOCK(inp); tp = intotcpcb(inp); CTR4(KTR_CXGBE, "%s: tid %u (%s), toep_flags 0x%x", __func__, tid, tp ? tcpstates[tp->t_state] : "no tp", toep->flags); if (toep->flags & TPF_ABORT_SHUTDOWN) goto done; so = inp->inp_socket; tp->snd_una = be32toh(cpl->snd_nxt) - 1; /* exclude FIN */ switch (tp->t_state) { case TCPS_CLOSING: /* see TCPS_FIN_WAIT_2 in do_peer_close too */ tcp_twstart(tp); release: INP_UNLOCK_ASSERT(inp); /* safe, we have a ref on the inp */ INP_INFO_WUNLOCK(&V_tcbinfo); INP_WLOCK(inp); final_cpl_received(toep); /* no more CPLs expected */ return (0); case TCPS_LAST_ACK: if (tcp_close(tp)) INP_WUNLOCK(inp); goto release; case TCPS_FIN_WAIT_1: if (so->so_rcv.sb_state & SBS_CANTRCVMORE) soisdisconnected(so); tp->t_state = TCPS_FIN_WAIT_2; break; default: log(LOG_ERR, "%s: TID %u received CPL_CLOSE_CON_RPL in state %s\n", __func__, tid, tcpstates[tp->t_state]); } done: INP_WUNLOCK(inp); INP_INFO_WUNLOCK(&V_tcbinfo); return (0); } void send_abort_rpl(struct adapter *sc, struct sge_wrq *ofld_txq, int tid, int rst_status) { struct wrqe *wr; struct cpl_abort_rpl *cpl; wr = alloc_wrqe(sizeof(*cpl), ofld_txq); if (wr == NULL) { /* XXX */ panic("%s: allocation failure.", __func__); } cpl = wrtod(wr); INIT_TP_WR_MIT_CPL(cpl, CPL_ABORT_RPL, tid); cpl->cmd = rst_status; t4_wrq_tx(sc, wr); } static int abort_status_to_errno(struct tcpcb *tp, unsigned int abort_reason) { switch (abort_reason) { case CPL_ERR_BAD_SYN: case CPL_ERR_CONN_RESET: return (tp->t_state == TCPS_CLOSE_WAIT ? EPIPE : ECONNRESET); case CPL_ERR_XMIT_TIMEDOUT: case CPL_ERR_PERSIST_TIMEDOUT: case CPL_ERR_FINWAIT2_TIMEDOUT: case CPL_ERR_KEEPALIVE_TIMEDOUT: return (ETIMEDOUT); default: return (EIO); } } int cpl_not_handled(struct sge_iq *, const struct rss_header *, struct mbuf *); /* * tom_cpl_iscsi_callback - * iscsi and tom would share the following cpl messages, so when any of these * message is received, after tom is done with processing it, the messages * needs to be forwarded to iscsi for further processing: * - CPL_SET_TCB_RPL * - CPL_RX_DATA_DDP */ void (*tom_cpl_iscsi_callback)(struct tom_data *, struct socket *, void *, unsigned int); struct mbuf *(*tom_queue_iscsi_callback)(struct socket *, unsigned int, int *); /* * Check if the handler function is set for a given CPL * return 0 if the function is NULL or cpl_not_handled, 1 otherwise. */ int t4tom_cpl_handler_registered(struct adapter *sc, unsigned int opcode) { MPASS(opcode < nitems(sc->cpl_handler)); return (sc->cpl_handler[opcode] && sc->cpl_handler[opcode] != cpl_not_handled); } /* * set the tom_cpl_iscsi_callback function, this function should be used * whenever both toe and iscsi need to process the same cpl msg. */ void t4tom_register_cpl_iscsi_callback(void (*fp)(struct tom_data *, struct socket *, void *, unsigned int)) { tom_cpl_iscsi_callback = fp; } void t4tom_register_queue_iscsi_callback(struct mbuf *(*fp)(struct socket *, unsigned int, int *qlen)) { tom_queue_iscsi_callback = fp; } int t4_cpl_iscsi_callback(struct tom_data *td, struct toepcb *toep, void *m, unsigned int opcode) { struct socket *so; if (opcode == CPL_FW4_ACK) so = toep->inp->inp_socket; else { INP_WLOCK(toep->inp); so = toep->inp->inp_socket; INP_WUNLOCK(toep->inp); } if (tom_cpl_iscsi_callback && so) { if (toep->ulp_mode == ULP_MODE_ISCSI) { tom_cpl_iscsi_callback(td, so, m, opcode); return (0); } } return (1); } struct mbuf * t4_queue_iscsi_callback(struct socket *so, struct toepcb *toep, unsigned int cmd, int *qlen) { if (tom_queue_iscsi_callback && so) { if (toep->ulp_mode == ULP_MODE_ISCSI) return (tom_queue_iscsi_callback(so, cmd, qlen)); } return (NULL); } /* * TCP RST from the peer, timeout, or some other such critical error. */ static int do_abort_req(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_abort_req_rss *cpl = (const void *)(rss + 1); unsigned int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct sge_wrq *ofld_txq = toep->ofld_txq; struct inpcb *inp; struct tcpcb *tp; #ifdef INVARIANTS unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); #endif KASSERT(opcode == CPL_ABORT_REQ_RSS, ("%s: unexpected opcode 0x%x", __func__, opcode)); KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); if (toep->flags & TPF_SYNQE) return (do_abort_req_synqe(iq, rss, m)); KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); if (negative_advice(cpl->status)) { CTR4(KTR_CXGBE, "%s: negative advice %d for tid %d (0x%x)", __func__, cpl->status, tid, toep->flags); return (0); /* Ignore negative advice */ } inp = toep->inp; INP_INFO_WLOCK(&V_tcbinfo); /* for tcp_close */ INP_WLOCK(inp); tp = intotcpcb(inp); CTR6(KTR_CXGBE, "%s: tid %d (%s), toep_flags 0x%x, inp_flags 0x%x, status %d", __func__, tid, tp ? tcpstates[tp->t_state] : "no tp", toep->flags, inp->inp_flags, cpl->status); /* * If we'd initiated an abort earlier the reply to it is responsible for * cleaning up resources. Otherwise we tear everything down right here * right now. We owe the T4 a CPL_ABORT_RPL no matter what. */ if (toep->flags & TPF_ABORT_SHUTDOWN) { INP_WUNLOCK(inp); goto done; } toep->flags |= TPF_ABORT_SHUTDOWN; if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) == 0) { struct socket *so = inp->inp_socket; if (so != NULL) so_error_set(so, abort_status_to_errno(tp, cpl->status)); tp = tcp_close(tp); if (tp == NULL) INP_WLOCK(inp); /* re-acquire */ } final_cpl_received(toep); done: INP_INFO_WUNLOCK(&V_tcbinfo); send_abort_rpl(sc, ofld_txq, tid, CPL_ABORT_NO_RST); return (0); } /* * Reply to the CPL_ABORT_REQ (send_reset) */ static int do_abort_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_abort_rpl_rss *cpl = (const void *)(rss + 1); unsigned int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct inpcb *inp = toep->inp; #ifdef INVARIANTS unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); #endif KASSERT(opcode == CPL_ABORT_RPL_RSS, ("%s: unexpected opcode 0x%x", __func__, opcode)); KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); if (toep->flags & TPF_SYNQE) return (do_abort_rpl_synqe(iq, rss, m)); KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); CTR5(KTR_CXGBE, "%s: tid %u, toep %p, inp %p, status %d", __func__, tid, toep, inp, cpl->status); KASSERT(toep->flags & TPF_ABORT_SHUTDOWN, ("%s: wasn't expecting abort reply", __func__)); INP_WLOCK(inp); final_cpl_received(toep); return (0); } static int do_rx_data(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_rx_data *cpl = mtod(m, const void *); unsigned int tid = GET_TID(cpl); struct toepcb *toep = lookup_tid(sc, tid); struct inpcb *inp = toep->inp; struct tcpcb *tp; struct socket *so; struct sockbuf *sb; int len; uint32_t ddp_placed = 0; if (__predict_false(toep->flags & TPF_SYNQE)) { #ifdef INVARIANTS struct synq_entry *synqe = (void *)toep; INP_WLOCK(synqe->lctx->inp); if (synqe->flags & TPF_SYNQE_HAS_L2TE) { KASSERT(synqe->flags & TPF_ABORT_SHUTDOWN, ("%s: listen socket closed but tid %u not aborted.", __func__, tid)); } else { /* * do_pass_accept_req is still running and will * eventually take care of this tid. */ } INP_WUNLOCK(synqe->lctx->inp); #endif CTR4(KTR_CXGBE, "%s: tid %u, synqe %p (0x%x)", __func__, tid, toep, toep->flags); m_freem(m); return (0); } KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); /* strip off CPL header */ m_adj(m, sizeof(*cpl)); len = m->m_pkthdr.len; INP_WLOCK(inp); if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) { CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), inp_flags 0x%x", __func__, tid, len, inp->inp_flags); INP_WUNLOCK(inp); m_freem(m); return (0); } tp = intotcpcb(inp); if (__predict_false(tp->rcv_nxt != be32toh(cpl->seq))) ddp_placed = be32toh(cpl->seq) - tp->rcv_nxt; tp->rcv_nxt += len; KASSERT(tp->rcv_wnd >= len, ("%s: negative window size", __func__)); tp->rcv_wnd -= len; tp->t_rcvtime = ticks; so = inp_inpcbtosocket(inp); sb = &so->so_rcv; SOCKBUF_LOCK(sb); if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) { CTR3(KTR_CXGBE, "%s: tid %u, excess rx (%d bytes)", __func__, tid, len); m_freem(m); SOCKBUF_UNLOCK(sb); INP_WUNLOCK(inp); INP_INFO_WLOCK(&V_tcbinfo); INP_WLOCK(inp); tp = tcp_drop(tp, ECONNRESET); if (tp) INP_WUNLOCK(inp); INP_INFO_WUNLOCK(&V_tcbinfo); return (0); } /* receive buffer autosize */ if (sb->sb_flags & SB_AUTOSIZE && V_tcp_do_autorcvbuf && sb->sb_hiwat < V_tcp_autorcvbuf_max && len > (sbspace(sb) / 8 * 7)) { unsigned int hiwat = sb->sb_hiwat; unsigned int newsize = min(hiwat + V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); if (!sbreserve_locked(sb, newsize, so, NULL)) sb->sb_flags &= ~SB_AUTOSIZE; else toep->rx_credits += newsize - hiwat; } if (toep->ulp_mode == ULP_MODE_TCPDDP) { int changed = !(toep->ddp_flags & DDP_ON) ^ cpl->ddp_off; if (changed) { if (toep->ddp_flags & DDP_SC_REQ) toep->ddp_flags ^= DDP_ON | DDP_SC_REQ; else { KASSERT(cpl->ddp_off == 1, ("%s: DDP switched on by itself.", __func__)); /* Fell out of DDP mode */ toep->ddp_flags &= ~(DDP_ON | DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE); if (ddp_placed) insert_ddp_data(toep, ddp_placed); } } if ((toep->ddp_flags & DDP_OK) == 0 && time_uptime >= toep->ddp_disabled + DDP_RETRY_WAIT) { toep->ddp_score = DDP_LOW_SCORE; toep->ddp_flags |= DDP_OK; CTR3(KTR_CXGBE, "%s: tid %u DDP_OK @ %u", __func__, tid, time_uptime); } if (toep->ddp_flags & DDP_ON) { /* * CPL_RX_DATA with DDP on can only be an indicate. Ask * soreceive to post a buffer or disable DDP. The * payload that arrived in this indicate is appended to * the socket buffer as usual. */ #if 0 CTR5(KTR_CXGBE, "%s: tid %u (0x%x) DDP indicate (seq 0x%x, len %d)", __func__, tid, toep->flags, be32toh(cpl->seq), len); #endif sb->sb_flags |= SB_DDP_INDICATE; } else if ((toep->ddp_flags & (DDP_OK|DDP_SC_REQ)) == DDP_OK && tp->rcv_wnd > DDP_RSVD_WIN && len >= sc->tt.ddp_thres) { /* * DDP allowed but isn't on (and a request to switch it * on isn't pending either), and conditions are ripe for * it to work. Switch it on. */ enable_ddp(sc, toep); } } KASSERT(toep->sb_cc >= sb->sb_cc, ("%s: sb %p has more data (%d) than last time (%d).", __func__, sb, sb->sb_cc, toep->sb_cc)); toep->rx_credits += toep->sb_cc - sb->sb_cc; sbappendstream_locked(sb, m); toep->sb_cc = sb->sb_cc; sorwakeup_locked(so); SOCKBUF_UNLOCK_ASSERT(sb); INP_WUNLOCK(inp); return (0); } #define S_CPL_FW4_ACK_OPCODE 24 #define M_CPL_FW4_ACK_OPCODE 0xff #define V_CPL_FW4_ACK_OPCODE(x) ((x) << S_CPL_FW4_ACK_OPCODE) #define G_CPL_FW4_ACK_OPCODE(x) \ (((x) >> S_CPL_FW4_ACK_OPCODE) & M_CPL_FW4_ACK_OPCODE) #define S_CPL_FW4_ACK_FLOWID 0 #define M_CPL_FW4_ACK_FLOWID 0xffffff #define V_CPL_FW4_ACK_FLOWID(x) ((x) << S_CPL_FW4_ACK_FLOWID) #define G_CPL_FW4_ACK_FLOWID(x) \ (((x) >> S_CPL_FW4_ACK_FLOWID) & M_CPL_FW4_ACK_FLOWID) #define S_CPL_FW4_ACK_CR 24 #define M_CPL_FW4_ACK_CR 0xff #define V_CPL_FW4_ACK_CR(x) ((x) << S_CPL_FW4_ACK_CR) #define G_CPL_FW4_ACK_CR(x) (((x) >> S_CPL_FW4_ACK_CR) & M_CPL_FW4_ACK_CR) #define S_CPL_FW4_ACK_SEQVAL 0 #define M_CPL_FW4_ACK_SEQVAL 0x1 #define V_CPL_FW4_ACK_SEQVAL(x) ((x) << S_CPL_FW4_ACK_SEQVAL) #define G_CPL_FW4_ACK_SEQVAL(x) \ (((x) >> S_CPL_FW4_ACK_SEQVAL) & M_CPL_FW4_ACK_SEQVAL) #define F_CPL_FW4_ACK_SEQVAL V_CPL_FW4_ACK_SEQVAL(1U) static int do_fw4_ack(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_fw4_ack *cpl = (const void *)(rss + 1); unsigned int tid = G_CPL_FW4_ACK_FLOWID(be32toh(OPCODE_TID(cpl))); struct toepcb *toep = lookup_tid(sc, tid); struct inpcb *inp; struct tcpcb *tp; struct socket *so; uint8_t credits = cpl->credits; struct ofld_tx_sdesc *txsd; int plen; #ifdef INVARIANTS unsigned int opcode = G_CPL_FW4_ACK_OPCODE(be32toh(OPCODE_TID(cpl))); #endif /* * Very unusual case: we'd sent a flowc + abort_req for a synq entry and * now this comes back carrying the credits for the flowc. */ if (__predict_false(toep->flags & TPF_SYNQE)) { KASSERT(toep->flags & TPF_ABORT_SHUTDOWN, ("%s: credits for a synq entry %p", __func__, toep)); return (0); } inp = toep->inp; KASSERT(opcode == CPL_FW4_ACK, ("%s: unexpected opcode 0x%x", __func__, opcode)); KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); INP_WLOCK(inp); if (__predict_false(toep->flags & TPF_ABORT_SHUTDOWN)) { INP_WUNLOCK(inp); return (0); } KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0, ("%s: inp_flags 0x%x", __func__, inp->inp_flags)); tp = intotcpcb(inp); if (cpl->flags & CPL_FW4_ACK_FLAGS_SEQVAL) { tcp_seq snd_una = be32toh(cpl->snd_una); #ifdef INVARIANTS if (__predict_false(SEQ_LT(snd_una, tp->snd_una))) { log(LOG_ERR, "%s: unexpected seq# %x for TID %u, snd_una %x\n", __func__, snd_una, toep->tid, tp->snd_una); } #endif if (tp->snd_una != snd_una) { tp->snd_una = snd_una; tp->ts_recent_age = tcp_ts_getticks(); } } so = inp->inp_socket; txsd = &toep->txsd[toep->txsd_cidx]; plen = 0; while (credits) { KASSERT(credits >= txsd->tx_credits, ("%s: too many (or partial) credits", __func__)); credits -= txsd->tx_credits; toep->tx_credits += txsd->tx_credits; plen += txsd->plen; txsd++; toep->txsd_avail++; KASSERT(toep->txsd_avail <= toep->txsd_total, ("%s: txsd avail > total", __func__)); if (__predict_false(++toep->txsd_cidx == toep->txsd_total)) { txsd = &toep->txsd[0]; toep->txsd_cidx = 0; } } if (toep->tx_credits == toep->tx_total) { toep->tx_nocompl = 0; toep->plen_nocompl = 0; } if (toep->flags & TPF_TX_SUSPENDED && toep->tx_credits >= toep->tx_total / 4) { toep->flags &= ~TPF_TX_SUSPENDED; if (toep->ulp_mode == ULP_MODE_ISCSI) t4_ulp_push_frames(sc, toep, plen); else t4_push_frames(sc, toep, plen); } else if (plen > 0) { struct sockbuf *sb = &so->so_snd; if (toep->ulp_mode == ULP_MODE_ISCSI) t4_cpl_iscsi_callback(toep->td, toep, &plen, CPL_FW4_ACK); else { SOCKBUF_LOCK(sb); sbdrop_locked(sb, plen); sowwakeup_locked(so); SOCKBUF_UNLOCK_ASSERT(sb); } } INP_WUNLOCK(inp); return (0); } static int do_set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_set_tcb_rpl *cpl = (const void *)(rss + 1); unsigned int tid = GET_TID(cpl); #ifdef INVARIANTS unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); #endif KASSERT(opcode == CPL_SET_TCB_RPL, ("%s: unexpected opcode 0x%x", __func__, opcode)); KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); if (is_ftid(sc, tid)) return (t4_filter_rpl(iq, rss, m)); /* TCB is a filter */ else { struct toepcb *toep = lookup_tid(sc, tid); t4_cpl_iscsi_callback(toep->td, toep, m, CPL_SET_TCB_RPL); return (0); } CXGBE_UNIMPLEMENTED(__func__); } void t4_set_tcb_field(struct adapter *sc, struct toepcb *toep, int ctrl, uint16_t word, uint64_t mask, uint64_t val) { struct wrqe *wr; struct cpl_set_tcb_field *req; wr = alloc_wrqe(sizeof(*req), ctrl ? toep->ctrlq : toep->ofld_txq); if (wr == NULL) { /* XXX */ panic("%s: allocation failure.", __func__); } req = wrtod(wr); INIT_TP_WR_MIT_CPL(req, CPL_SET_TCB_FIELD, toep->tid); req->reply_ctrl = htobe16(V_NO_REPLY(1) | V_QUEUENO(toep->ofld_rxq->iq.abs_id)); req->word_cookie = htobe16(V_WORD(word) | V_COOKIE(0)); req->mask = htobe64(mask); req->val = htobe64(val); t4_wrq_tx(sc, wr); } void t4_init_cpl_io_handlers(struct adapter *sc) { t4_register_cpl_handler(sc, CPL_PEER_CLOSE, do_peer_close); t4_register_cpl_handler(sc, CPL_CLOSE_CON_RPL, do_close_con_rpl); t4_register_cpl_handler(sc, CPL_ABORT_REQ_RSS, do_abort_req); t4_register_cpl_handler(sc, CPL_ABORT_RPL_RSS, do_abort_rpl); t4_register_cpl_handler(sc, CPL_RX_DATA, do_rx_data); t4_register_cpl_handler(sc, CPL_FW4_ACK, do_fw4_ack); t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, do_set_tcb_rpl); } void t4_uninit_cpl_io_handlers(struct adapter *sc) { t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, t4_filter_rpl); } #endif Index: stable/10 =================================================================== --- stable/10 (revision 283855) +++ stable/10 (revision 283856) Property changes on: stable/10 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r273480,273750,273753,273797,274461