diff --git a/sys/dev/ntb/if_ntb/if_ntb.c b/sys/dev/ntb/if_ntb/if_ntb.c index 169296564d74..8b1fa7a5bd63 100644 --- a/sys/dev/ntb/if_ntb/if_ntb.c +++ b/sys/dev/ntb/if_ntb/if_ntb.c @@ -1,1457 +1,1464 @@ /*- * Copyright (C) 2013 Intel Corporation * All rights reserved. * * 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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "../ntb_hw/ntb_hw.h" /* * The Non-Transparent Bridge (NTB) is a device on some Intel processors that * allows you to connect two systems using a PCI-e link. * * This module contains a protocol for sending and receiving messages, and * exposes that protocol through a simulated ethernet device called ntb. * * NOTE: Much of the code in this module is shared with Linux. Any patches may * be picked up and redistributed in Linux with a dual GPL/BSD license. */ /* TODO: These functions should really be part of the kernel */ #define test_bit(pos, bitmap_addr) (*(bitmap_addr) & 1UL << (pos)) #define set_bit(pos, bitmap_addr) *(bitmap_addr) |= 1UL << (pos) #define clear_bit(pos, bitmap_addr) *(bitmap_addr) &= ~(1UL << (pos)) #define KTR_NTB KTR_SPARE3 #define NTB_TRANSPORT_VERSION 3 #define NTB_RX_MAX_PKTS 64 #define NTB_RXQ_SIZE 300 static unsigned int transport_mtu = 0x4000 + ETHER_HDR_LEN + ETHER_CRC_LEN; -/* - * This is an oversimplification to work around Xeon Errata. The second client - * may be usable for unidirectional traffic. - */ -static unsigned int max_num_clients = 1; +static unsigned int max_num_clients; +SYSCTL_UINT(_hw_ntb, OID_AUTO, max_num_clients, CTLFLAG_RDTUN, + &max_num_clients, 0, "Maximum number of NTB transport clients. " + "0 (default) - use all available NTB memory windows; " + "positive integer N - Limit to N memory windows."); STAILQ_HEAD(ntb_queue_list, ntb_queue_entry); struct ntb_queue_entry { /* ntb_queue list reference */ STAILQ_ENTRY(ntb_queue_entry) entry; /* info on data to be transfered */ void *cb_data; void *buf; uint64_t len; uint64_t flags; }; struct ntb_rx_info { unsigned int entry; }; struct ntb_transport_qp { struct ntb_netdev *transport; struct ntb_softc *ntb; void *cb_data; bool client_ready; bool qp_link; uint8_t qp_num; /* Only 64 QPs are allowed. 0-63 */ struct ntb_rx_info *rx_info; struct ntb_rx_info *remote_rx_info; void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data, void *data, int len); struct ntb_queue_list tx_free_q; struct mtx ntb_tx_free_q_lock; void *tx_mw; uint64_t tx_index; uint64_t tx_max_entry; uint64_t tx_max_frame; void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data, void *data, int len); struct ntb_queue_list rx_pend_q; struct ntb_queue_list rx_free_q; struct mtx ntb_rx_pend_q_lock; struct mtx ntb_rx_free_q_lock; struct task rx_completion_task; void *rx_buff; uint64_t rx_index; uint64_t rx_max_entry; uint64_t rx_max_frame; void (*event_handler)(void *data, enum ntb_link_event status); struct callout link_work; struct callout queue_full; struct callout rx_full; uint64_t last_rx_no_buf; /* Stats */ uint64_t rx_bytes; uint64_t rx_pkts; uint64_t rx_ring_empty; uint64_t rx_err_no_buf; uint64_t rx_err_oflow; uint64_t rx_err_ver; uint64_t tx_bytes; uint64_t tx_pkts; uint64_t tx_ring_full; }; struct ntb_queue_handlers { void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data, void *data, int len); void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data, void *data, int len); void (*event_handler)(void *data, enum ntb_link_event status); }; struct ntb_transport_mw { size_t size; void *virt_addr; vm_paddr_t dma_addr; }; struct ntb_netdev { struct ntb_softc *ntb; struct ifnet *ifp; - struct ntb_transport_mw mw[NTB_NUM_MW]; + struct ntb_transport_mw mw[NTB_MAX_NUM_MW]; struct ntb_transport_qp *qps; uint64_t max_qps; uint64_t qp_bitmap; bool transport_link; struct callout link_work; struct ntb_transport_qp *qp; uint64_t bufsize; u_char eaddr[ETHER_ADDR_LEN]; struct mtx tx_lock; struct mtx rx_lock; }; static struct ntb_netdev net_softc; enum { IF_NTB_DESC_DONE_FLAG = 1 << 0, IF_NTB_LINK_DOWN_FLAG = 1 << 1, }; struct ntb_payload_header { uint64_t ver; uint64_t len; uint64_t flags; }; enum { /* * The order of this enum is part of the if_ntb remote protocol. Do * not reorder without bumping protocol version (and it's probably best * to keep the protocol in lock-step with the Linux NTB driver. */ IF_NTB_VERSION = 0, IF_NTB_QP_LINKS, IF_NTB_NUM_QPS, IF_NTB_NUM_MWS, /* * N.B.: transport_link_work assumes MW1 enums = MW0 + 2. */ IF_NTB_MW0_SZ_HIGH, IF_NTB_MW0_SZ_LOW, IF_NTB_MW1_SZ_HIGH, IF_NTB_MW1_SZ_LOW, IF_NTB_MAX_SPAD, }; -#define QP_TO_MW(qp) ((qp) % NTB_NUM_MW) +#define QP_TO_MW(ntb, qp) ((qp) % ntb_get_max_mw(ntb)) #define NTB_QP_DEF_NUM_ENTRIES 100 #define NTB_LINK_DOWN_TIMEOUT 10 static int ntb_handle_module_events(struct module *m, int what, void *arg); static int ntb_setup_interface(void); static int ntb_teardown_interface(void); static void ntb_net_init(void *arg); static int ntb_ioctl(struct ifnet *ifp, u_long command, caddr_t data); static void ntb_start(struct ifnet *ifp); static void ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data, int len); static void ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data, int len); static void ntb_net_event_handler(void *data, enum ntb_link_event status); static int ntb_transport_init(struct ntb_softc *ntb); static void ntb_transport_free(void *transport); static void ntb_transport_init_queue(struct ntb_netdev *nt, unsigned int qp_num); static void ntb_transport_free_queue(struct ntb_transport_qp *qp); static struct ntb_transport_qp * ntb_transport_create_queue(void *data, struct ntb_softc *pdev, const struct ntb_queue_handlers *handlers); static void ntb_transport_link_up(struct ntb_transport_qp *qp); static int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, unsigned int len); static int ntb_process_tx(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry); static void ntb_tx_copy_task(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry, void *offset); static void ntb_qp_full(void *arg); static int ntb_transport_rxc_db(void *arg, int dummy); static void ntb_rx_pendq_full(void *arg); static int ntb_process_rxc(struct ntb_transport_qp *qp); static void ntb_rx_copy_task(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry, void *offset); static void ntb_rx_completion_task(void *arg, int pending); static void ntb_transport_event_callback(void *data, enum ntb_hw_event event); static void ntb_transport_link_work(void *arg); static int ntb_set_mw(struct ntb_netdev *nt, int num_mw, unsigned int size); static void ntb_free_mw(struct ntb_netdev *nt, int num_mw); static void ntb_transport_setup_qp_mw(struct ntb_netdev *nt, unsigned int qp_num); static void ntb_qp_link_work(void *arg); static void ntb_transport_link_cleanup(struct ntb_netdev *nt); static void ntb_qp_link_down(struct ntb_transport_qp *qp); static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp); static void ntb_transport_link_down(struct ntb_transport_qp *qp); static void ntb_send_link_down(struct ntb_transport_qp *qp); static void ntb_list_add(struct mtx *lock, struct ntb_queue_entry *entry, struct ntb_queue_list *list); static struct ntb_queue_entry *ntb_list_rm(struct mtx *lock, struct ntb_queue_list *list); static void create_random_local_eui48(u_char *eaddr); static unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp); MALLOC_DEFINE(M_NTB_IF, "if_ntb", "ntb network driver"); /* Module setup and teardown */ static int ntb_handle_module_events(struct module *m, int what, void *arg) { int err = 0; switch (what) { case MOD_LOAD: err = ntb_setup_interface(); break; case MOD_UNLOAD: err = ntb_teardown_interface(); break; default: err = EOPNOTSUPP; break; } return (err); } static moduledata_t if_ntb_mod = { "if_ntb", ntb_handle_module_events, NULL }; DECLARE_MODULE(if_ntb, if_ntb_mod, SI_SUB_KLD, SI_ORDER_ANY); MODULE_DEPEND(if_ntb, ntb_hw, 1, 1, 1); static int ntb_setup_interface(void) { struct ifnet *ifp; struct ntb_queue_handlers handlers = { ntb_net_rx_handler, ntb_net_tx_handler, ntb_net_event_handler }; net_softc.ntb = devclass_get_softc(devclass_find("ntb_hw"), 0); if (net_softc.ntb == NULL) { printf("ntb: Cannot find devclass\n"); return (ENXIO); } ntb_transport_init(net_softc.ntb); ifp = net_softc.ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { printf("ntb: cannot allocate ifnet structure\n"); return (ENOMEM); } net_softc.qp = ntb_transport_create_queue(ifp, net_softc.ntb, &handlers); if_initname(ifp, "ntb", 0); ifp->if_init = ntb_net_init; ifp->if_softc = &net_softc; ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX; ifp->if_ioctl = ntb_ioctl; ifp->if_start = ntb_start; IFQ_SET_MAXLEN(&ifp->if_snd, IFQ_MAXLEN); ifp->if_snd.ifq_drv_maxlen = IFQ_MAXLEN; IFQ_SET_READY(&ifp->if_snd); create_random_local_eui48(net_softc.eaddr); ether_ifattach(ifp, net_softc.eaddr); ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_JUMBO_MTU; ifp->if_capenable = ifp->if_capabilities; ntb_transport_link_up(net_softc.qp); net_softc.bufsize = ntb_transport_max_size(net_softc.qp) + sizeof(struct ether_header); return (0); } static int ntb_teardown_interface(void) { if (net_softc.qp != NULL) ntb_transport_link_down(net_softc.qp); if (net_softc.ifp != NULL) { ether_ifdetach(net_softc.ifp); if_free(net_softc.ifp); } if (net_softc.qp != NULL) { ntb_transport_free_queue(net_softc.qp); ntb_transport_free(&net_softc); } return (0); } /* Network device interface */ static void ntb_net_init(void *arg) { struct ntb_netdev *ntb_softc = arg; struct ifnet *ifp = ntb_softc->ifp; ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; ifp->if_flags |= IFF_UP; if_link_state_change(ifp, LINK_STATE_UP); } static int ntb_ioctl(struct ifnet *ifp, u_long command, caddr_t data) { struct ntb_netdev *nt = ifp->if_softc; struct ifreq *ifr = (struct ifreq *)data; int error = 0; switch (command) { case SIOCSIFMTU: { if (ifr->ifr_mtu > ntb_transport_max_size(nt->qp) - ETHER_HDR_LEN - ETHER_CRC_LEN) { error = EINVAL; break; } ifp->if_mtu = ifr->ifr_mtu; break; } default: error = ether_ioctl(ifp, command, data); break; } return (error); } static void ntb_start(struct ifnet *ifp) { struct mbuf *m_head; struct ntb_netdev *nt = ifp->if_softc; int rc; mtx_lock(&nt->tx_lock); ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; CTR0(KTR_NTB, "TX: ntb_start"); while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) { IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head); CTR1(KTR_NTB, "TX: start mbuf %p", m_head); rc = ntb_transport_tx_enqueue(nt->qp, m_head, m_head, m_length(m_head, NULL)); if (rc != 0) { CTR1(KTR_NTB, "TX: could not tx mbuf %p. Returning to snd q", m_head); if (rc == EAGAIN) { ifp->if_drv_flags |= IFF_DRV_OACTIVE; IFQ_DRV_PREPEND(&ifp->if_snd, m_head); callout_reset(&nt->qp->queue_full, hz / 1000, ntb_qp_full, ifp); } break; } } mtx_unlock(&nt->tx_lock); } /* Network Device Callbacks */ static void ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data, int len) { m_freem(data); CTR1(KTR_NTB, "TX: tx_handler freeing mbuf %p", data); } static void ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data, int len) { struct mbuf *m = data; struct ifnet *ifp = qp_data; CTR0(KTR_NTB, "RX: rx handler"); (*ifp->if_input)(ifp, m); } static void ntb_net_event_handler(void *data, enum ntb_link_event status) { struct ifnet *ifp; ifp = data; (void)ifp; /* XXX The Linux driver munges with the carrier status here. */ switch (status) { case NTB_LINK_DOWN: break; case NTB_LINK_UP: break; default: panic("Bogus ntb_link_event %u\n", status); } } /* Transport Init and teardown */ static int ntb_transport_init(struct ntb_softc *ntb) { struct ntb_netdev *nt = &net_softc; int rc, i; - nt->max_qps = max_num_clients; + if (max_num_clients == 0) + nt->max_qps = MIN(ntb_get_max_cbs(ntb), ntb_get_max_mw(ntb)); + else + nt->max_qps = MIN(ntb_get_max_cbs(ntb), max_num_clients); + ntb_register_transport(ntb, nt); mtx_init(&nt->tx_lock, "ntb transport tx", NULL, MTX_DEF); mtx_init(&nt->rx_lock, "ntb transport rx", NULL, MTX_DEF); nt->qps = malloc(nt->max_qps * sizeof(struct ntb_transport_qp), M_NTB_IF, M_WAITOK|M_ZERO); nt->qp_bitmap = ((uint64_t) 1 << nt->max_qps) - 1; for (i = 0; i < nt->max_qps; i++) ntb_transport_init_queue(nt, i); callout_init(&nt->link_work, 0); rc = ntb_register_event_callback(ntb, ntb_transport_event_callback); if (rc != 0) goto err; if (ntb_query_link_status(ntb)) { if (bootverbose) device_printf(ntb_get_device(ntb), "link up\n"); callout_reset(&nt->link_work, 0, ntb_transport_link_work, nt); } return (0); err: free(nt->qps, M_NTB_IF); ntb_unregister_transport(ntb); return (rc); } static void ntb_transport_free(void *transport) { struct ntb_netdev *nt = transport; struct ntb_softc *ntb = nt->ntb; int i; ntb_transport_link_cleanup(nt); callout_drain(&nt->link_work); /* verify that all the qps are freed */ for (i = 0; i < nt->max_qps; i++) if (!test_bit(i, &nt->qp_bitmap)) ntb_transport_free_queue(&nt->qps[i]); ntb_unregister_event_callback(ntb); - for (i = 0; i < NTB_NUM_MW; i++) + for (i = 0; i < NTB_MAX_NUM_MW; i++) ntb_free_mw(nt, i); free(nt->qps, M_NTB_IF); ntb_unregister_transport(ntb); } static void ntb_transport_init_queue(struct ntb_netdev *nt, unsigned int qp_num) { struct ntb_transport_qp *qp; unsigned int num_qps_mw, tx_size; - uint8_t mw_num = QP_TO_MW(qp_num); + uint8_t mw_num, mw_max; + + mw_max = ntb_get_max_mw(nt->ntb); + mw_num = QP_TO_MW(nt->ntb, qp_num); qp = &nt->qps[qp_num]; qp->qp_num = qp_num; qp->transport = nt; qp->ntb = nt->ntb; qp->qp_link = NTB_LINK_DOWN; qp->client_ready = NTB_LINK_DOWN; qp->event_handler = NULL; - if (nt->max_qps % NTB_NUM_MW && mw_num + 1 < nt->max_qps / NTB_NUM_MW) - num_qps_mw = nt->max_qps / NTB_NUM_MW + 1; + if (nt->max_qps % mw_max && mw_num + 1 < nt->max_qps / mw_max) + num_qps_mw = nt->max_qps / mw_max + 1; else - num_qps_mw = nt->max_qps / NTB_NUM_MW; + num_qps_mw = nt->max_qps / mw_max; tx_size = (unsigned int) ntb_get_mw_size(qp->ntb, mw_num) / num_qps_mw; qp->rx_info = (struct ntb_rx_info *) ((char *)ntb_get_mw_vbase(qp->ntb, mw_num) + - (qp_num / NTB_NUM_MW * tx_size)); + (qp_num / mw_max * tx_size)); tx_size -= sizeof(struct ntb_rx_info); qp->tx_mw = qp->rx_info + 1; /* Due to house-keeping, there must be at least 2 buffs */ qp->tx_max_frame = min(transport_mtu + sizeof(struct ntb_payload_header), tx_size / 2); qp->tx_max_entry = tx_size / qp->tx_max_frame; callout_init(&qp->link_work, 0); callout_init(&qp->queue_full, 1); callout_init(&qp->rx_full, 1); mtx_init(&qp->ntb_rx_pend_q_lock, "ntb rx pend q", NULL, MTX_SPIN); mtx_init(&qp->ntb_rx_free_q_lock, "ntb rx free q", NULL, MTX_SPIN); mtx_init(&qp->ntb_tx_free_q_lock, "ntb tx free q", NULL, MTX_SPIN); TASK_INIT(&qp->rx_completion_task, 0, ntb_rx_completion_task, qp); STAILQ_INIT(&qp->rx_pend_q); STAILQ_INIT(&qp->rx_free_q); STAILQ_INIT(&qp->tx_free_q); } static void ntb_transport_free_queue(struct ntb_transport_qp *qp) { struct ntb_queue_entry *entry; if (qp == NULL) return; callout_drain(&qp->link_work); ntb_unregister_db_callback(qp->ntb, qp->qp_num); while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q))) free(entry, M_NTB_IF); while ((entry = ntb_list_rm(&qp->ntb_rx_pend_q_lock, &qp->rx_pend_q))) free(entry, M_NTB_IF); while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) free(entry, M_NTB_IF); set_bit(qp->qp_num, &qp->transport->qp_bitmap); } /** * ntb_transport_create_queue - Create a new NTB transport layer queue * @rx_handler: receive callback function * @tx_handler: transmit callback function * @event_handler: event callback function * * Create a new NTB transport layer queue and provide the queue with a callback * routine for both transmit and receive. The receive callback routine will be * used to pass up data when the transport has received it on the queue. The * transmit callback routine will be called when the transport has completed the * transmission of the data on the queue and the data is ready to be freed. * * RETURNS: pointer to newly created ntb_queue, NULL on error. */ static struct ntb_transport_qp * ntb_transport_create_queue(void *data, struct ntb_softc *pdev, const struct ntb_queue_handlers *handlers) { struct ntb_queue_entry *entry; struct ntb_transport_qp *qp; struct ntb_netdev *nt; unsigned int free_queue; int rc, i; nt = ntb_find_transport(pdev); if (nt == NULL) goto err; free_queue = ffs(nt->qp_bitmap); if (free_queue == 0) goto err; /* decrement free_queue to make it zero based */ free_queue--; clear_bit(free_queue, &nt->qp_bitmap); qp = &nt->qps[free_queue]; qp->cb_data = data; qp->rx_handler = handlers->rx_handler; qp->tx_handler = handlers->tx_handler; qp->event_handler = handlers->event_handler; for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) { entry = malloc(sizeof(struct ntb_queue_entry), M_NTB_IF, M_WAITOK|M_ZERO); entry->cb_data = nt->ifp; entry->buf = NULL; entry->len = transport_mtu; ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q); } for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) { entry = malloc(sizeof(struct ntb_queue_entry), M_NTB_IF, M_WAITOK|M_ZERO); ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q); } rc = ntb_register_db_callback(qp->ntb, free_queue, qp, ntb_transport_rxc_db); if (rc != 0) goto err1; return (qp); err1: while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) free(entry, M_NTB_IF); while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q))) free(entry, M_NTB_IF); set_bit(free_queue, &nt->qp_bitmap); err: return (NULL); } /** * ntb_transport_link_up - Notify NTB transport of client readiness to use queue * @qp: NTB transport layer queue to be enabled * * Notify NTB transport layer of client readiness to use queue */ static void ntb_transport_link_up(struct ntb_transport_qp *qp) { if (qp == NULL) return; qp->client_ready = NTB_LINK_UP; if (bootverbose) device_printf(ntb_get_device(qp->ntb), "qp client ready\n"); if (qp->transport->transport_link == NTB_LINK_UP) callout_reset(&qp->link_work, 0, ntb_qp_link_work, qp); } /* Transport Tx */ /** * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry * @qp: NTB transport layer queue the entry is to be enqueued on * @cb: per buffer pointer for callback function to use * @data: pointer to data buffer that will be sent * @len: length of the data buffer * * Enqueue a new transmit buffer onto the transport queue from which a NTB * payload will be transmitted. This assumes that a lock is being held to * serialize access to the qp. * * RETURNS: An appropriate ERRNO error value on error, or zero for success. */ static int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, unsigned int len) { struct ntb_queue_entry *entry; int rc; if (qp == NULL || qp->qp_link != NTB_LINK_UP || len == 0) { CTR0(KTR_NTB, "TX: link not up"); return (EINVAL); } entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); if (entry == NULL) { CTR0(KTR_NTB, "TX: could not get entry from tx_free_q"); return (ENOMEM); } CTR1(KTR_NTB, "TX: got entry %p from tx_free_q", entry); entry->cb_data = cb; entry->buf = data; entry->len = len; entry->flags = 0; rc = ntb_process_tx(qp, entry); if (rc != 0) { ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q); CTR1(KTR_NTB, "TX: process_tx failed. Returning entry %p to tx_free_q", entry); } return (rc); } static int ntb_process_tx(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry) { void *offset; offset = (char *)qp->tx_mw + qp->tx_max_frame * qp->tx_index; CTR3(KTR_NTB, "TX: process_tx: tx_pkts=%u, tx_index=%u, remote entry=%u", qp->tx_pkts, qp->tx_index, qp->remote_rx_info->entry); if (qp->tx_index == qp->remote_rx_info->entry) { CTR0(KTR_NTB, "TX: ring full"); qp->tx_ring_full++; return (EAGAIN); } if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) { if (qp->tx_handler != NULL) qp->tx_handler(qp, qp->cb_data, entry->buf, EIO); ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q); CTR1(KTR_NTB, "TX: frame too big. returning entry %p to tx_free_q", entry); return (0); } CTR2(KTR_NTB, "TX: copying entry %p to offset %p", entry, offset); ntb_tx_copy_task(qp, entry, offset); qp->tx_index++; qp->tx_index %= qp->tx_max_entry; qp->tx_pkts++; return (0); } static void ntb_tx_copy_task(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry, void *offset) { struct ntb_payload_header *hdr; CTR2(KTR_NTB, "TX: copying %d bytes to offset %p", entry->len, offset); if (entry->buf != NULL) m_copydata((struct mbuf *)entry->buf, 0, entry->len, offset); hdr = (struct ntb_payload_header *)((char *)offset + qp->tx_max_frame - sizeof(struct ntb_payload_header)); hdr->len = entry->len; /* TODO: replace with bus_space_write */ hdr->ver = qp->tx_pkts; /* TODO: replace with bus_space_write */ wmb(); /* TODO: replace with bus_space_write */ hdr->flags = entry->flags | IF_NTB_DESC_DONE_FLAG; ntb_ring_doorbell(qp->ntb, qp->qp_num); /* * The entry length can only be zero if the packet is intended to be a * "link down" or similar. Since no payload is being sent in these * cases, there is nothing to add to the completion queue. */ if (entry->len > 0) { qp->tx_bytes += entry->len; if (qp->tx_handler) qp->tx_handler(qp, qp->cb_data, entry->cb_data, entry->len); } CTR2(KTR_NTB, "TX: entry %p sent. hdr->ver = %d, Returning to tx_free_q", entry, hdr->ver); ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q); } static void ntb_qp_full(void *arg) { CTR0(KTR_NTB, "TX: qp_full callout"); ntb_start(arg); } /* Transport Rx */ static void ntb_rx_pendq_full(void *arg) { CTR0(KTR_NTB, "RX: ntb_rx_pendq_full callout"); ntb_transport_rxc_db(arg, 0); } static int ntb_transport_rxc_db(void *arg, int dummy __unused) { struct ntb_transport_qp *qp = arg; uint64_t i; int rc; /* * Limit the number of packets processed in a single interrupt to * provide fairness to others */ mtx_lock(&qp->transport->rx_lock); CTR0(KTR_NTB, "RX: transport_rx"); for (i = 0; i < MIN(qp->rx_max_entry, INT_MAX); i++) { rc = ntb_process_rxc(qp); if (rc != 0) { CTR0(KTR_NTB, "RX: process_rxc failed"); break; } } mtx_unlock(&qp->transport->rx_lock); return ((int)i); } static int ntb_process_rxc(struct ntb_transport_qp *qp) { struct ntb_payload_header *hdr; struct ntb_queue_entry *entry; void *offset; offset = (void *) ((char *)qp->rx_buff + qp->rx_max_frame * qp->rx_index); hdr = (void *) ((char *)offset + qp->rx_max_frame - sizeof(struct ntb_payload_header)); CTR1(KTR_NTB, "RX: process_rxc rx_index = %u", qp->rx_index); entry = ntb_list_rm(&qp->ntb_rx_pend_q_lock, &qp->rx_pend_q); if (entry == NULL) { qp->rx_err_no_buf++; CTR0(KTR_NTB, "RX: No entries in rx_pend_q"); return (ENOMEM); } callout_stop(&qp->rx_full); CTR1(KTR_NTB, "RX: rx entry %p from rx_pend_q", entry); if ((hdr->flags & IF_NTB_DESC_DONE_FLAG) == 0) { CTR1(KTR_NTB, "RX: hdr not done. Returning entry %p to rx_pend_q", entry); ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q); qp->rx_ring_empty++; return (EAGAIN); } if (hdr->ver != (uint32_t) qp->rx_pkts) { CTR3(KTR_NTB,"RX: ver != rx_pkts (%x != %lx). " "Returning entry %p to rx_pend_q", hdr->ver, qp->rx_pkts, entry); ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q); qp->rx_err_ver++; return (EIO); } if ((hdr->flags & IF_NTB_LINK_DOWN_FLAG) != 0) { ntb_qp_link_down(qp); CTR1(KTR_NTB, "RX: link down. adding entry %p back to rx_pend_q", entry); ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q); goto out; } if (hdr->len <= entry->len) { entry->len = hdr->len; ntb_rx_copy_task(qp, entry, offset); } else { CTR1(KTR_NTB, "RX: len too long. Returning entry %p to rx_pend_q", entry); ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q); qp->rx_err_oflow++; } qp->rx_bytes += hdr->len; qp->rx_pkts++; CTR1(KTR_NTB, "RX: received %ld rx_pkts", qp->rx_pkts); out: /* Ensure that the data is globally visible before clearing the flag */ wmb(); hdr->flags = 0; /* TODO: replace with bus_space_write */ qp->rx_info->entry = qp->rx_index; qp->rx_index++; qp->rx_index %= qp->rx_max_entry; return (0); } static void ntb_rx_copy_task(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry, void *offset) { struct ifnet *ifp = entry->cb_data; unsigned int len = entry->len; struct mbuf *m; CTR2(KTR_NTB, "RX: copying %d bytes from offset %p", len, offset); m = m_devget(offset, len, 0, ifp, NULL); m->m_pkthdr.csum_flags = CSUM_IP_CHECKED | CSUM_IP_VALID; entry->buf = (void *)m; CTR2(KTR_NTB, "RX: copied entry %p to mbuf %p. Adding entry to rx_free_q", entry, m); ntb_list_add(&qp->ntb_rx_free_q_lock, entry, &qp->rx_free_q); taskqueue_enqueue(taskqueue_swi, &qp->rx_completion_task); } static void ntb_rx_completion_task(void *arg, int pending) { struct ntb_transport_qp *qp = arg; struct mbuf *m; struct ntb_queue_entry *entry; CTR0(KTR_NTB, "RX: rx_completion_task"); while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q))) { m = entry->buf; CTR2(KTR_NTB, "RX: completing entry %p, mbuf %p", entry, m); if (qp->rx_handler && qp->client_ready == NTB_LINK_UP) qp->rx_handler(qp, qp->cb_data, m, entry->len); entry->buf = NULL; entry->len = qp->transport->bufsize; CTR1(KTR_NTB,"RX: entry %p removed from rx_free_q " "and added to rx_pend_q", entry); ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q); if (qp->rx_err_no_buf > qp->last_rx_no_buf) { qp->last_rx_no_buf = qp->rx_err_no_buf; CTR0(KTR_NTB, "RX: could spawn rx task"); callout_reset(&qp->rx_full, hz / 1000, ntb_rx_pendq_full, qp); } } } /* Link Event handler */ static void ntb_transport_event_callback(void *data, enum ntb_hw_event event) { struct ntb_netdev *nt = data; switch (event) { case NTB_EVENT_HW_LINK_UP: if (bootverbose) device_printf(ntb_get_device(nt->ntb), "HW link up\n"); callout_reset(&nt->link_work, 0, ntb_transport_link_work, nt); break; case NTB_EVENT_HW_LINK_DOWN: if (bootverbose) device_printf(ntb_get_device(nt->ntb), "HW link down\n"); ntb_transport_link_cleanup(nt); break; default: panic("ntb: Unknown NTB event"); } } /* Link bring up */ static void ntb_transport_link_work(void *arg) { struct ntb_netdev *nt = arg; struct ntb_softc *ntb = nt->ntb; struct ntb_transport_qp *qp; uint64_t val64; uint32_t val, i, num_mw; int rc; - if (ntb_has_feature(ntb, NTB_REGS_THRU_MW)) - num_mw = NTB_NUM_MW - 1; - else - num_mw = NTB_NUM_MW; + num_mw = ntb_get_max_mw(ntb); /* send the local info, in the opposite order of the way we read it */ for (i = 0; i < num_mw; i++) { rc = ntb_write_remote_spad(ntb, IF_NTB_MW0_SZ_HIGH + (i * 2), (uint64_t)ntb_get_mw_size(ntb, i) >> 32); if (rc != 0) goto out; rc = ntb_write_remote_spad(ntb, IF_NTB_MW0_SZ_LOW + (i * 2), (uint32_t)ntb_get_mw_size(ntb, i)); if (rc != 0) goto out; } rc = ntb_write_remote_spad(ntb, IF_NTB_NUM_MWS, num_mw); if (rc != 0) goto out; rc = ntb_write_remote_spad(ntb, IF_NTB_NUM_QPS, nt->max_qps); if (rc != 0) goto out; rc = ntb_write_remote_spad(ntb, IF_NTB_VERSION, NTB_TRANSPORT_VERSION); if (rc != 0) goto out; /* Query the remote side for its info */ rc = ntb_read_local_spad(ntb, IF_NTB_VERSION, &val); if (rc != 0) goto out; if (val != NTB_TRANSPORT_VERSION) goto out; rc = ntb_read_local_spad(ntb, IF_NTB_NUM_QPS, &val); if (rc != 0) goto out; if (val != nt->max_qps) goto out; rc = ntb_read_local_spad(ntb, IF_NTB_NUM_MWS, &val); if (rc != 0) goto out; if (val != num_mw) goto out; for (i = 0; i < num_mw; i++) { rc = ntb_read_local_spad(ntb, IF_NTB_MW0_SZ_HIGH + (i * 2), &val); if (rc != 0) goto free_mws; val64 = (uint64_t)val << 32; rc = ntb_read_local_spad(ntb, IF_NTB_MW0_SZ_LOW + (i * 2), &val); if (rc != 0) goto free_mws; val64 |= val; rc = ntb_set_mw(nt, i, val64); if (rc != 0) goto free_mws; } nt->transport_link = NTB_LINK_UP; if (bootverbose) device_printf(ntb_get_device(ntb), "transport link up\n"); for (i = 0; i < nt->max_qps; i++) { qp = &nt->qps[i]; ntb_transport_setup_qp_mw(nt, i); if (qp->client_ready == NTB_LINK_UP) callout_reset(&qp->link_work, 0, ntb_qp_link_work, qp); } return; free_mws: - for (i = 0; i < NTB_NUM_MW; i++) + for (i = 0; i < NTB_MAX_NUM_MW; i++) ntb_free_mw(nt, i); out: if (ntb_query_link_status(ntb)) callout_reset(&nt->link_work, NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_transport_link_work, nt); } static int ntb_set_mw(struct ntb_netdev *nt, int num_mw, unsigned int size) { struct ntb_transport_mw *mw = &nt->mw[num_mw]; /* No need to re-setup */ if (mw->size == size) return (0); if (mw->size != 0) ntb_free_mw(nt, num_mw); /* Alloc memory for receiving data. Must be 4k aligned */ mw->size = size; mw->virt_addr = contigmalloc(mw->size, M_NTB_IF, M_ZERO, 0, BUS_SPACE_MAXADDR, mw->size, 0); if (mw->virt_addr == NULL) { mw->size = 0; printf("ntb: Unable to allocate MW buffer of size %zu\n", mw->size); return (ENOMEM); } /* TODO: replace with bus_space_* functions */ mw->dma_addr = vtophys(mw->virt_addr); /* * Ensure that the allocation from contigmalloc is aligned as * requested. XXX: This may not be needed -- brought in for parity * with the Linux driver. */ if (mw->dma_addr % size != 0) { device_printf(ntb_get_device(nt->ntb), "DMA memory 0x%jx not aligned to BAR size 0x%x\n", (uintmax_t)mw->dma_addr, size); ntb_free_mw(nt, num_mw); return (ENOMEM); } /* Notify HW the memory location of the receive buffer */ ntb_set_mw_addr(nt->ntb, num_mw, mw->dma_addr); return (0); } static void ntb_free_mw(struct ntb_netdev *nt, int num_mw) { struct ntb_transport_mw *mw = &nt->mw[num_mw]; if (mw->virt_addr == NULL) return; contigfree(mw->virt_addr, mw->size, M_NTB_IF); mw->virt_addr = NULL; } static void ntb_transport_setup_qp_mw(struct ntb_netdev *nt, unsigned int qp_num) { struct ntb_transport_qp *qp = &nt->qps[qp_num]; void *offset; unsigned int rx_size, num_qps_mw; - uint8_t mw_num = QP_TO_MW(qp_num); + uint8_t mw_num, mw_max; unsigned int i; - if (nt->max_qps % NTB_NUM_MW && mw_num + 1 < nt->max_qps / NTB_NUM_MW) - num_qps_mw = nt->max_qps / NTB_NUM_MW + 1; + mw_max = ntb_get_max_mw(nt->ntb); + mw_num = QP_TO_MW(nt->ntb, qp_num); + + if (nt->max_qps % mw_max && mw_num + 1 < nt->max_qps / mw_max) + num_qps_mw = nt->max_qps / mw_max + 1; else - num_qps_mw = nt->max_qps / NTB_NUM_MW; + num_qps_mw = nt->max_qps / mw_max; rx_size = (unsigned int) nt->mw[mw_num].size / num_qps_mw; qp->remote_rx_info = (void *)((uint8_t *)nt->mw[mw_num].virt_addr + - (qp_num / NTB_NUM_MW * rx_size)); + (qp_num / mw_max * rx_size)); rx_size -= sizeof(struct ntb_rx_info); qp->rx_buff = qp->remote_rx_info + 1; /* Due to house-keeping, there must be at least 2 buffs */ qp->rx_max_frame = min(transport_mtu + sizeof(struct ntb_payload_header), rx_size / 2); qp->rx_max_entry = rx_size / qp->rx_max_frame; qp->rx_index = 0; qp->remote_rx_info->entry = qp->rx_max_entry - 1; /* setup the hdr offsets with 0's */ for (i = 0; i < qp->rx_max_entry; i++) { offset = (void *)((uint8_t *)qp->rx_buff + qp->rx_max_frame * (i + 1) - sizeof(struct ntb_payload_header)); memset(offset, 0, sizeof(struct ntb_payload_header)); } qp->rx_pkts = 0; qp->tx_pkts = 0; qp->tx_index = 0; } static void ntb_qp_link_work(void *arg) { struct ntb_transport_qp *qp = arg; struct ntb_softc *ntb = qp->ntb; struct ntb_netdev *nt = qp->transport; int rc, val; rc = ntb_read_remote_spad(ntb, IF_NTB_QP_LINKS, &val); if (rc != 0) return; rc = ntb_write_remote_spad(ntb, IF_NTB_QP_LINKS, val | 1 << qp->qp_num); /* query remote spad for qp ready bits */ rc = ntb_read_local_spad(ntb, IF_NTB_QP_LINKS, &val); /* See if the remote side is up */ if ((1 << qp->qp_num & val) != 0) { qp->qp_link = NTB_LINK_UP; if (qp->event_handler != NULL) qp->event_handler(qp->cb_data, NTB_LINK_UP); if (bootverbose) device_printf(ntb_get_device(ntb), "qp link up\n"); } else if (nt->transport_link == NTB_LINK_UP) { callout_reset(&qp->link_work, NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_qp_link_work, qp); } } /* Link down event*/ static void ntb_transport_link_cleanup(struct ntb_netdev *nt) { int i; /* Pass along the info to any clients */ for (i = 0; i < nt->max_qps; i++) if (!test_bit(i, &nt->qp_bitmap)) ntb_qp_link_down(&nt->qps[i]); if (nt->transport_link == NTB_LINK_DOWN) callout_drain(&nt->link_work); else nt->transport_link = NTB_LINK_DOWN; /* * The scratchpad registers keep the values if the remote side * goes down, blast them now to give them a sane value the next * time they are accessed */ for (i = 0; i < IF_NTB_MAX_SPAD; i++) ntb_write_local_spad(nt->ntb, i, 0); } static void ntb_qp_link_down(struct ntb_transport_qp *qp) { ntb_qp_link_cleanup(qp); } static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp) { struct ntb_netdev *nt = qp->transport; if (qp->qp_link == NTB_LINK_DOWN) { callout_drain(&qp->link_work); return; } if (qp->event_handler != NULL) qp->event_handler(qp->cb_data, NTB_LINK_DOWN); qp->qp_link = NTB_LINK_DOWN; if (nt->transport_link == NTB_LINK_UP) callout_reset(&qp->link_work, NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_qp_link_work, qp); } /* Link commanded down */ /** * ntb_transport_link_down - Notify NTB transport to no longer enqueue data * @qp: NTB transport layer queue to be disabled * * Notify NTB transport layer of client's desire to no longer receive data on * transport queue specified. It is the client's responsibility to ensure all * entries on queue are purged or otherwise handled appropriately. */ static void ntb_transport_link_down(struct ntb_transport_qp *qp) { int rc, val; if (qp == NULL) return; qp->client_ready = NTB_LINK_DOWN; rc = ntb_read_remote_spad(qp->ntb, IF_NTB_QP_LINKS, &val); if (rc != 0) return; rc = ntb_write_remote_spad(qp->ntb, IF_NTB_QP_LINKS, val & ~(1 << qp->qp_num)); if (qp->qp_link == NTB_LINK_UP) ntb_send_link_down(qp); else callout_drain(&qp->link_work); } static void ntb_send_link_down(struct ntb_transport_qp *qp) { struct ntb_queue_entry *entry; int i, rc; if (qp->qp_link == NTB_LINK_DOWN) return; qp->qp_link = NTB_LINK_DOWN; for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) { entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); if (entry != NULL) break; pause("NTB Wait for link down", hz / 10); } if (entry == NULL) return; entry->cb_data = NULL; entry->buf = NULL; entry->len = 0; entry->flags = IF_NTB_LINK_DOWN_FLAG; mtx_lock(&qp->transport->tx_lock); rc = ntb_process_tx(qp, entry); if (rc != 0) printf("ntb: Failed to send link down\n"); mtx_unlock(&qp->transport->tx_lock); } /* List Management */ static void ntb_list_add(struct mtx *lock, struct ntb_queue_entry *entry, struct ntb_queue_list *list) { mtx_lock_spin(lock); STAILQ_INSERT_TAIL(list, entry, entry); mtx_unlock_spin(lock); } static struct ntb_queue_entry * ntb_list_rm(struct mtx *lock, struct ntb_queue_list *list) { struct ntb_queue_entry *entry; mtx_lock_spin(lock); if (STAILQ_EMPTY(list)) { entry = NULL; goto out; } entry = STAILQ_FIRST(list); STAILQ_REMOVE_HEAD(list, entry); out: mtx_unlock_spin(lock); return (entry); } /* Helper functions */ /* TODO: This too should really be part of the kernel */ #define EUI48_MULTICAST 1 << 0 #define EUI48_LOCALLY_ADMINISTERED 1 << 1 static void create_random_local_eui48(u_char *eaddr) { static uint8_t counter = 0; uint32_t seed = ticks; eaddr[0] = EUI48_LOCALLY_ADMINISTERED; memcpy(&eaddr[1], &seed, sizeof(uint32_t)); eaddr[5] = counter++; } /** * ntb_transport_max_size - Query the max payload size of a qp * @qp: NTB transport layer queue to be queried * * Query the maximum payload size permissible on the given qp * * RETURNS: the max payload size of a qp */ static unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp) { if (qp == NULL) return (0); return (qp->tx_max_frame - sizeof(struct ntb_payload_header)); } diff --git a/sys/dev/ntb/ntb_hw/ntb_hw.c b/sys/dev/ntb/ntb_hw/ntb_hw.c index 58ff611ff7a0..c91a1e2ad43e 100644 --- a/sys/dev/ntb/ntb_hw/ntb_hw.c +++ b/sys/dev/ntb/ntb_hw/ntb_hw.c @@ -1,1789 +1,1827 @@ /*- * Copyright (C) 2013 Intel Corporation * All rights reserved. * * 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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "ntb_regs.h" #include "ntb_hw.h" /* * The Non-Transparent Bridge (NTB) is a device on some Intel processors that * allows you to connect two systems using a PCI-e link. * * This module contains the hardware abstraction layer for the NTB. It allows * you to send and recieve interrupts, map the memory windows and send and * receive messages in the scratch-pad registers. * * NOTE: Much of the code in this module is shared with Linux. Any patches may * be picked up and redistributed in Linux with a dual GPL/BSD license. */ #define NTB_CONFIG_BAR 0 #define NTB_B2B_BAR_1 1 #define NTB_B2B_BAR_2 2 #define NTB_MAX_BARS 3 #define NTB_MW_TO_BAR(mw) ((mw) + 1) #define MAX_MSIX_INTERRUPTS MAX(XEON_MAX_DB_BITS, SOC_MAX_DB_BITS) #define NTB_HB_TIMEOUT 1 /* second */ #define SOC_LINK_RECOVERY_TIME 500 #define DEVICE2SOFTC(dev) ((struct ntb_softc *) device_get_softc(dev)) enum ntb_device_type { NTB_XEON, NTB_SOC }; /* Device features and workarounds */ #define HAS_FEATURE(feature) \ ((ntb->features & (feature)) != 0) struct ntb_hw_info { uint32_t device_id; const char *desc; enum ntb_device_type type; uint64_t features; }; struct ntb_pci_bar_info { bus_space_tag_t pci_bus_tag; bus_space_handle_t pci_bus_handle; int pci_resource_id; struct resource *pci_resource; vm_paddr_t pbase; void *vbase; u_long size; }; struct ntb_int_info { struct resource *res; int rid; void *tag; }; struct ntb_db_cb { ntb_db_callback callback; unsigned int db_num; void *data; struct ntb_softc *ntb; struct callout irq_work; bool reserved; }; struct ntb_softc { device_t device; enum ntb_device_type type; uint64_t features; struct ntb_pci_bar_info bar_info[NTB_MAX_BARS]; struct ntb_int_info int_info[MAX_MSIX_INTERRUPTS]; uint32_t allocated_interrupts; struct callout heartbeat_timer; struct callout lr_timer; void *ntb_transport; ntb_event_callback event_cb; - struct ntb_db_cb *db_cb; + struct ntb_db_cb *db_cb; + uint8_t max_cbs; struct { + uint8_t max_mw; uint8_t max_spads; uint8_t max_db_bits; uint8_t msix_cnt; } limits; struct { uint32_t ldb; uint32_t ldb_mask; uint32_t rdb; uint32_t bar2_xlat; uint32_t bar4_xlat; uint32_t spad_remote; uint32_t spad_local; uint32_t lnk_cntl; uint32_t lnk_stat; uint32_t spci_cmd; } reg_ofs; uint32_t ppd; uint8_t conn_type; uint8_t dev_type; uint8_t bits_per_vector; uint8_t link_status; uint8_t link_width; uint8_t link_speed; }; #ifdef __i386__ static __inline uint64_t bus_space_read_8(bus_space_tag_t tag, bus_space_handle_t handle, bus_size_t offset) { return (bus_space_read_4(tag, handle, offset) | ((uint64_t)bus_space_read_4(tag, handle, offset + 4)) << 32); } static __inline void bus_space_write_8(bus_space_tag_t tag, bus_space_handle_t handle, bus_size_t offset, uint64_t val) { bus_space_write_4(tag, handle, offset, val); bus_space_write_4(tag, handle, offset + 4, val >> 32); } #endif #define ntb_bar_read(SIZE, bar, offset) \ bus_space_read_ ## SIZE (ntb->bar_info[(bar)].pci_bus_tag, \ ntb->bar_info[(bar)].pci_bus_handle, (offset)) #define ntb_bar_write(SIZE, bar, offset, val) \ bus_space_write_ ## SIZE (ntb->bar_info[(bar)].pci_bus_tag, \ ntb->bar_info[(bar)].pci_bus_handle, (offset), (val)) #define ntb_reg_read(SIZE, offset) ntb_bar_read(SIZE, NTB_CONFIG_BAR, offset) #define ntb_reg_write(SIZE, offset, val) \ ntb_bar_write(SIZE, NTB_CONFIG_BAR, offset, val) #define ntb_mw_read(SIZE, offset) ntb_bar_read(SIZE, NTB_B2B_BAR_2, offset) #define ntb_mw_write(SIZE, offset, val) \ ntb_bar_write(SIZE, NTB_B2B_BAR_2, offset, val) typedef int (*bar_map_strategy)(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar); static int ntb_probe(device_t device); static int ntb_attach(device_t device); static int ntb_detach(device_t device); static int ntb_map_pci_bars(struct ntb_softc *ntb); static int map_pci_bar(struct ntb_softc *ntb, bar_map_strategy strategy, struct ntb_pci_bar_info *bar); static int map_mmr_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar); static int map_memory_window_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar); static void ntb_unmap_pci_bar(struct ntb_softc *ntb); static int ntb_remap_msix(device_t, uint32_t desired, uint32_t avail); static int ntb_setup_interrupts(struct ntb_softc *ntb); static int ntb_setup_legacy_interrupt(struct ntb_softc *ntb); static int ntb_setup_xeon_msix(struct ntb_softc *ntb, uint32_t num_vectors); static int ntb_setup_soc_msix(struct ntb_softc *ntb, uint32_t num_vectors); static void ntb_teardown_interrupts(struct ntb_softc *ntb); static void handle_soc_irq(void *arg); static void handle_xeon_irq(void *arg); static void handle_xeon_event_irq(void *arg); static void ntb_handle_legacy_interrupt(void *arg); static void ntb_irq_work(void *arg); static uint64_t db_ioread(struct ntb_softc *, uint32_t regoff); static void db_iowrite(struct ntb_softc *, uint32_t regoff, uint64_t val); static void mask_ldb_interrupt(struct ntb_softc *ntb, unsigned int idx); static void unmask_ldb_interrupt(struct ntb_softc *ntb, unsigned int idx); static int ntb_create_callbacks(struct ntb_softc *ntb, uint32_t num_vectors); static void ntb_free_callbacks(struct ntb_softc *ntb); static struct ntb_hw_info *ntb_get_device_info(uint32_t device_id); static int ntb_detect_xeon(struct ntb_softc *ntb); static int ntb_detect_soc(struct ntb_softc *ntb); static int ntb_setup_xeon(struct ntb_softc *ntb); static int ntb_setup_soc(struct ntb_softc *ntb); static void ntb_teardown_xeon(struct ntb_softc *ntb); static void configure_soc_secondary_side_bars(struct ntb_softc *ntb); static void configure_xeon_secondary_side_bars(struct ntb_softc *ntb); static void ntb_handle_heartbeat(void *arg); static void ntb_handle_link_event(struct ntb_softc *ntb, int link_state); static void ntb_hw_link_down(struct ntb_softc *ntb); static void ntb_hw_link_up(struct ntb_softc *ntb); static void recover_soc_link(void *arg); static int ntb_check_link_status(struct ntb_softc *ntb); static void save_bar_parameters(struct ntb_pci_bar_info *bar); static struct ntb_hw_info pci_ids[] = { { 0x0C4E8086, "Atom Processor S1200 NTB Primary B2B", NTB_SOC, 0 }, /* XXX: PS/SS IDs left out until they are supported. */ { 0x37258086, "JSF Xeon C35xx/C55xx Non-Transparent Bridge B2B", NTB_XEON, NTB_REGS_THRU_MW | NTB_B2BDOORBELL_BIT14 }, { 0x3C0D8086, "SNB Xeon E5/Core i7 Non-Transparent Bridge B2B", NTB_XEON, NTB_REGS_THRU_MW | NTB_B2BDOORBELL_BIT14 }, { 0x0E0D8086, "IVT Xeon E5 V2 Non-Transparent Bridge B2B", NTB_XEON, NTB_REGS_THRU_MW | NTB_B2BDOORBELL_BIT14 | NTB_SB01BASE_LOCKUP | NTB_BAR_SIZE_4K }, { 0x2F0D8086, "HSX Xeon E5 V3 Non-Transparent Bridge B2B", NTB_XEON, NTB_REGS_THRU_MW | NTB_B2BDOORBELL_BIT14 | NTB_SB01BASE_LOCKUP }, { 0x6F0D8086, "BDX Xeon E5 V4 Non-Transparent Bridge B2B", NTB_XEON, NTB_REGS_THRU_MW | NTB_B2BDOORBELL_BIT14 | NTB_SB01BASE_LOCKUP }, { 0x00000000, NULL, NTB_SOC, 0 } }; /* * OS <-> Driver interface structures */ MALLOC_DEFINE(M_NTB, "ntb_hw", "ntb_hw driver memory allocations"); static device_method_t ntb_pci_methods[] = { /* Device interface */ DEVMETHOD(device_probe, ntb_probe), DEVMETHOD(device_attach, ntb_attach), DEVMETHOD(device_detach, ntb_detach), DEVMETHOD_END }; static driver_t ntb_pci_driver = { "ntb_hw", ntb_pci_methods, sizeof(struct ntb_softc), }; static devclass_t ntb_devclass; DRIVER_MODULE(ntb_hw, pci, ntb_pci_driver, ntb_devclass, NULL, NULL); MODULE_VERSION(ntb_hw, 1); SYSCTL_NODE(_hw, OID_AUTO, ntb, CTLFLAG_RW, 0, "NTB sysctls"); /* * OS <-> Driver linkage functions */ static int ntb_probe(device_t device) { struct ntb_hw_info *p; p = ntb_get_device_info(pci_get_devid(device)); if (p == NULL) return (ENXIO); device_set_desc(device, p->desc); return (0); } static int ntb_attach(device_t device) { struct ntb_softc *ntb; struct ntb_hw_info *p; int error; ntb = DEVICE2SOFTC(device); p = ntb_get_device_info(pci_get_devid(device)); ntb->device = device; ntb->type = p->type; ntb->features = p->features; /* Heartbeat timer for NTB_SOC since there is no link interrupt */ callout_init(&ntb->heartbeat_timer, 1); callout_init(&ntb->lr_timer, 1); if (ntb->type == NTB_SOC) error = ntb_detect_soc(ntb); else error = ntb_detect_xeon(ntb); if (error) goto out; + ntb->limits.max_mw = NTB_MAX_NUM_MW; + error = ntb_map_pci_bars(ntb); if (error) goto out; if (ntb->type == NTB_SOC) error = ntb_setup_soc(ntb); else error = ntb_setup_xeon(ntb); if (error) goto out; error = ntb_setup_interrupts(ntb); if (error) goto out; pci_enable_busmaster(ntb->device); out: if (error != 0) ntb_detach(device); return (error); } static int ntb_detach(device_t device) { struct ntb_softc *ntb; ntb = DEVICE2SOFTC(device); callout_drain(&ntb->heartbeat_timer); callout_drain(&ntb->lr_timer); if (ntb->type == NTB_XEON) ntb_teardown_xeon(ntb); ntb_teardown_interrupts(ntb); ntb_unmap_pci_bar(ntb); return (0); } static int ntb_map_pci_bars(struct ntb_softc *ntb) { int rc; ntb->bar_info[NTB_CONFIG_BAR].pci_resource_id = PCIR_BAR(0); rc = map_pci_bar(ntb, map_mmr_bar, &ntb->bar_info[NTB_CONFIG_BAR]); if (rc != 0) return (rc); ntb->bar_info[NTB_B2B_BAR_1].pci_resource_id = PCIR_BAR(2); rc = map_pci_bar(ntb, map_memory_window_bar, &ntb->bar_info[NTB_B2B_BAR_1]); if (rc != 0) return (rc); ntb->bar_info[NTB_B2B_BAR_2].pci_resource_id = PCIR_BAR(4); if (HAS_FEATURE(NTB_REGS_THRU_MW)) rc = map_pci_bar(ntb, map_mmr_bar, &ntb->bar_info[NTB_B2B_BAR_2]); else rc = map_pci_bar(ntb, map_memory_window_bar, &ntb->bar_info[NTB_B2B_BAR_2]); return (rc); } static int map_pci_bar(struct ntb_softc *ntb, bar_map_strategy strategy, struct ntb_pci_bar_info *bar) { int rc; rc = strategy(ntb, bar); if (rc != 0) device_printf(ntb->device, "unable to allocate pci resource\n"); else device_printf(ntb->device, "Bar size = %lx, v %p, p %p\n", bar->size, bar->vbase, (void *)(bar->pbase)); return (rc); } static int map_mmr_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar) { bar->pci_resource = bus_alloc_resource_any(ntb->device, SYS_RES_MEMORY, &bar->pci_resource_id, RF_ACTIVE); if (bar->pci_resource == NULL) return (ENXIO); save_bar_parameters(bar); return (0); } static int map_memory_window_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar) { int rc; uint8_t bar_size_bits = 0; bar->pci_resource = bus_alloc_resource_any(ntb->device, SYS_RES_MEMORY, &bar->pci_resource_id, RF_ACTIVE); if (bar->pci_resource == NULL) return (ENXIO); save_bar_parameters(bar); /* * Ivytown NTB BAR sizes are misreported by the hardware due to a * hardware issue. To work around this, query the size it should be * configured to by the device and modify the resource to correspond to * this new size. The BIOS on systems with this problem is required to * provide enough address space to allow the driver to make this change * safely. * * Ideally I could have just specified the size when I allocated the * resource like: * bus_alloc_resource(ntb->device, * SYS_RES_MEMORY, &bar->pci_resource_id, 0ul, ~0ul, * 1ul << bar_size_bits, RF_ACTIVE); * but the PCI driver does not honor the size in this call, so we have * to modify it after the fact. */ if (HAS_FEATURE(NTB_BAR_SIZE_4K)) { if (bar->pci_resource_id == PCIR_BAR(2)) bar_size_bits = pci_read_config(ntb->device, XEON_PBAR23SZ_OFFSET, 1); else bar_size_bits = pci_read_config(ntb->device, XEON_PBAR45SZ_OFFSET, 1); rc = bus_adjust_resource(ntb->device, SYS_RES_MEMORY, bar->pci_resource, bar->pbase, bar->pbase + (1ul << bar_size_bits) - 1); if (rc != 0) { device_printf(ntb->device, "unable to resize bar\n"); return (rc); } save_bar_parameters(bar); } /* Mark bar region as write combining to improve performance. */ rc = pmap_change_attr((vm_offset_t)bar->vbase, bar->size, VM_MEMATTR_WRITE_COMBINING); if (rc != 0) { device_printf(ntb->device, "unable to mark bar as WRITE_COMBINING\n"); return (rc); } return (0); } static void ntb_unmap_pci_bar(struct ntb_softc *ntb) { struct ntb_pci_bar_info *current_bar; int i; for (i = 0; i< NTB_MAX_BARS; i++) { current_bar = &ntb->bar_info[i]; if (current_bar->pci_resource != NULL) bus_release_resource(ntb->device, SYS_RES_MEMORY, current_bar->pci_resource_id, current_bar->pci_resource); } } static int ntb_setup_xeon_msix(struct ntb_softc *ntb, uint32_t num_vectors) { void (*interrupt_handler)(void *); void *int_arg; uint32_t i; int rc; if (num_vectors < 4) return (ENOSPC); for (i = 0; i < num_vectors; i++) { ntb->int_info[i].rid = i + 1; ntb->int_info[i].res = bus_alloc_resource_any(ntb->device, SYS_RES_IRQ, &ntb->int_info[i].rid, RF_ACTIVE); if (ntb->int_info[i].res == NULL) { device_printf(ntb->device, "bus_alloc_resource failed\n"); return (ENOMEM); } ntb->int_info[i].tag = NULL; ntb->allocated_interrupts++; if (i == num_vectors - 1) { interrupt_handler = handle_xeon_event_irq; int_arg = ntb; } else { interrupt_handler = handle_xeon_irq; int_arg = &ntb->db_cb[i]; } rc = bus_setup_intr(ntb->device, ntb->int_info[i].res, INTR_MPSAFE | INTR_TYPE_MISC, NULL, interrupt_handler, int_arg, &ntb->int_info[i].tag); if (rc != 0) { device_printf(ntb->device, "bus_setup_intr failed\n"); return (ENXIO); } } /* * Prevent consumers from registering callbacks on the link event irq * slot, from which they will never be called back. */ ntb->db_cb[num_vectors - 1].reserved = true; + ntb->max_cbs--; return (0); } static int ntb_setup_soc_msix(struct ntb_softc *ntb, uint32_t num_vectors) { uint32_t i; int rc; for (i = 0; i < num_vectors; i++) { ntb->int_info[i].rid = i + 1; ntb->int_info[i].res = bus_alloc_resource_any(ntb->device, SYS_RES_IRQ, &ntb->int_info[i].rid, RF_ACTIVE); if (ntb->int_info[i].res == NULL) { device_printf(ntb->device, "bus_alloc_resource failed\n"); return (ENOMEM); } ntb->int_info[i].tag = NULL; ntb->allocated_interrupts++; rc = bus_setup_intr(ntb->device, ntb->int_info[i].res, INTR_MPSAFE | INTR_TYPE_MISC, NULL, handle_soc_irq, &ntb->db_cb[i], &ntb->int_info[i].tag); if (rc != 0) { device_printf(ntb->device, "bus_setup_intr failed\n"); return (ENXIO); } } return (0); } /* * The Linux NTB driver drops from MSI-X to legacy INTx if a unique vector * cannot be allocated for each MSI-X message. JHB seems to think remapping * should be okay. This tunable should enable us to test that hypothesis * when someone gets their hands on some Xeon hardware. */ static int ntb_force_remap_mode; SYSCTL_INT(_hw_ntb, OID_AUTO, force_remap_mode, CTLFLAG_RDTUN, &ntb_force_remap_mode, 0, "If enabled, force MSI-X messages to be remapped" " to a smaller number of ithreads, even if the desired number are " "available"); /* * In case it is NOT ok, give consumers an abort button. */ static int ntb_prefer_intx; SYSCTL_INT(_hw_ntb, OID_AUTO, prefer_intx_to_remap, CTLFLAG_RDTUN, &ntb_prefer_intx, 0, "If enabled, prefer to use legacy INTx mode rather " "than remapping MSI-X messages over available slots (match Linux driver " "behavior)"); /* * Remap the desired number of MSI-X messages to available ithreads in a simple * round-robin fashion. */ static int ntb_remap_msix(device_t dev, uint32_t desired, uint32_t avail) { u_int *vectors; uint32_t i; int rc; if (ntb_prefer_intx != 0) return (ENXIO); vectors = malloc(desired * sizeof(*vectors), M_NTB, M_ZERO | M_WAITOK); for (i = 0; i < desired; i++) vectors[i] = (i % avail) + 1; rc = pci_remap_msix(dev, desired, vectors); free(vectors, M_NTB); return (rc); } static int ntb_setup_interrupts(struct ntb_softc *ntb) { uint32_t desired_vectors, num_vectors; uint64_t mask; int rc; ntb->allocated_interrupts = 0; /* * On SOC, disable all interrupts. On XEON, disable all but Link * Interrupt. The rest will be unmasked as callbacks are registered. */ mask = 0; if (ntb->type == NTB_XEON) mask = (1 << XEON_LINK_DB); db_iowrite(ntb, ntb->reg_ofs.ldb_mask, ~mask); num_vectors = desired_vectors = MIN(pci_msix_count(ntb->device), ntb->limits.max_db_bits); if (desired_vectors >= 1) { rc = pci_alloc_msix(ntb->device, &num_vectors); if (ntb_force_remap_mode != 0 && rc == 0 && num_vectors == desired_vectors) num_vectors--; if (rc == 0 && num_vectors < desired_vectors) { rc = ntb_remap_msix(ntb->device, desired_vectors, num_vectors); if (rc == 0) num_vectors = desired_vectors; else pci_release_msi(ntb->device); } if (rc != 0) num_vectors = 1; } else num_vectors = 1; + /* + * If allocating MSI-X interrupts succeeds, limit callbacks to the + * number of MSI-X slots available. + */ ntb_create_callbacks(ntb, num_vectors); if (ntb->type == NTB_XEON) rc = ntb_setup_xeon_msix(ntb, num_vectors); else rc = ntb_setup_soc_msix(ntb, num_vectors); - if (rc != 0) + if (rc != 0) { device_printf(ntb->device, "Error allocating MSI-X interrupts: %d\n", rc); + /* + * If allocating MSI-X interrupts failed and we're forced to + * use legacy INTx anyway, the only limit on individual + * callbacks is the number of doorbell bits. + * + * CEM: This seems odd to me but matches the behavior of the + * Linux driver ca. September 2013 + */ + ntb_free_callbacks(ntb); + ntb_create_callbacks(ntb, ntb->limits.max_db_bits); + } + if (ntb->type == NTB_XEON && rc == ENOSPC) rc = ntb_setup_legacy_interrupt(ntb); return (rc); } static int ntb_setup_legacy_interrupt(struct ntb_softc *ntb) { int rc; ntb->int_info[0].rid = 0; ntb->int_info[0].res = bus_alloc_resource_any(ntb->device, SYS_RES_IRQ, &ntb->int_info[0].rid, RF_SHAREABLE|RF_ACTIVE); if (ntb->int_info[0].res == NULL) { device_printf(ntb->device, "bus_alloc_resource failed\n"); return (ENOMEM); } ntb->int_info[0].tag = NULL; ntb->allocated_interrupts = 1; rc = bus_setup_intr(ntb->device, ntb->int_info[0].res, INTR_MPSAFE | INTR_TYPE_MISC, NULL, ntb_handle_legacy_interrupt, ntb, &ntb->int_info[0].tag); if (rc != 0) { device_printf(ntb->device, "bus_setup_intr failed\n"); return (ENXIO); } return (0); } static void ntb_teardown_interrupts(struct ntb_softc *ntb) { struct ntb_int_info *current_int; int i; for (i = 0; i < ntb->allocated_interrupts; i++) { current_int = &ntb->int_info[i]; if (current_int->tag != NULL) bus_teardown_intr(ntb->device, current_int->res, current_int->tag); if (current_int->res != NULL) bus_release_resource(ntb->device, SYS_RES_IRQ, rman_get_rid(current_int->res), current_int->res); } ntb_free_callbacks(ntb); pci_release_msi(ntb->device); } /* * Doorbell register and mask are 64-bit on SoC, 16-bit on Xeon. Abstract it * out to make code clearer. */ static uint64_t db_ioread(struct ntb_softc *ntb, uint32_t regoff) { if (ntb->type == NTB_SOC) return (ntb_reg_read(8, regoff)); KASSERT(ntb->type == NTB_XEON, ("bad ntb type")); return (ntb_reg_read(2, regoff)); } static void db_iowrite(struct ntb_softc *ntb, uint32_t regoff, uint64_t val) { if (ntb->type == NTB_SOC) { ntb_reg_write(8, regoff, val); return; } KASSERT(ntb->type == NTB_XEON, ("bad ntb type")); ntb_reg_write(2, regoff, (uint16_t)val); } static void mask_ldb_interrupt(struct ntb_softc *ntb, unsigned int idx) { uint64_t mask; mask = db_ioread(ntb, ntb->reg_ofs.ldb_mask); mask |= 1 << (idx * ntb->bits_per_vector); db_iowrite(ntb, ntb->reg_ofs.ldb_mask, mask); } static void unmask_ldb_interrupt(struct ntb_softc *ntb, unsigned int idx) { uint64_t mask; mask = db_ioread(ntb, ntb->reg_ofs.ldb_mask); mask &= ~(1 << (idx * ntb->bits_per_vector)); db_iowrite(ntb, ntb->reg_ofs.ldb_mask, mask); } static void handle_soc_irq(void *arg) { struct ntb_db_cb *db_cb = arg; struct ntb_softc *ntb = db_cb->ntb; db_iowrite(ntb, ntb->reg_ofs.ldb, (uint64_t) 1 << db_cb->db_num); if (db_cb->callback != NULL) { mask_ldb_interrupt(ntb, db_cb->db_num); callout_reset(&db_cb->irq_work, 0, ntb_irq_work, db_cb); } } static void handle_xeon_irq(void *arg) { struct ntb_db_cb *db_cb = arg; struct ntb_softc *ntb = db_cb->ntb; /* * On Xeon, there are 16 bits in the interrupt register * but only 4 vectors. So, 5 bits are assigned to the first 3 * vectors, with the 4th having a single bit for link * interrupts. */ db_iowrite(ntb, ntb->reg_ofs.ldb, ((1 << ntb->bits_per_vector) - 1) << (db_cb->db_num * ntb->bits_per_vector)); if (db_cb->callback != NULL) { mask_ldb_interrupt(ntb, db_cb->db_num); callout_reset(&db_cb->irq_work, 0, ntb_irq_work, db_cb); } } /* Since we do not have a HW doorbell in SOC, this is only used in JF/JT */ static void handle_xeon_event_irq(void *arg) { struct ntb_softc *ntb = arg; int rc; rc = ntb_check_link_status(ntb); if (rc != 0) device_printf(ntb->device, "Error determining link status\n"); /* bit 15 is always the link bit */ db_iowrite(ntb, ntb->reg_ofs.ldb, 1 << XEON_LINK_DB); } static void ntb_handle_legacy_interrupt(void *arg) { struct ntb_softc *ntb = arg; unsigned int i; uint64_t ldb; ldb = db_ioread(ntb, ntb->reg_ofs.ldb); if (ntb->type == NTB_XEON && (ldb & XEON_DB_HW_LINK) != 0) { handle_xeon_event_irq(ntb); ldb &= ~XEON_DB_HW_LINK; } while (ldb != 0) { i = ffs(ldb); ldb &= ldb - 1; if (ntb->type == NTB_SOC) handle_soc_irq(&ntb->db_cb[i]); else handle_xeon_irq(&ntb->db_cb[i]); } } static int ntb_create_callbacks(struct ntb_softc *ntb, uint32_t num_vectors) { uint32_t i; + ntb->max_cbs = num_vectors; ntb->db_cb = malloc(num_vectors * sizeof(*ntb->db_cb), M_NTB, M_ZERO | M_WAITOK); for (i = 0; i < num_vectors; i++) { ntb->db_cb[i].db_num = i; ntb->db_cb[i].ntb = ntb; } return (0); } static void ntb_free_callbacks(struct ntb_softc *ntb) { uint8_t i; - for (i = 0; i < ntb->limits.max_db_bits; i++) + for (i = 0; i < ntb->max_cbs; i++) ntb_unregister_db_callback(ntb, i); free(ntb->db_cb, M_NTB); + ntb->max_cbs = 0; } static struct ntb_hw_info * ntb_get_device_info(uint32_t device_id) { struct ntb_hw_info *ep = pci_ids; while (ep->device_id) { if (ep->device_id == device_id) return (ep); ++ep; } return (NULL); } static void ntb_teardown_xeon(struct ntb_softc *ntb) { ntb_hw_link_down(ntb); } static int ntb_detect_xeon(struct ntb_softc *ntb) { uint8_t ppd, conn_type; ppd = pci_read_config(ntb->device, NTB_PPD_OFFSET, 1); ntb->ppd = ppd; if ((ppd & XEON_PPD_DEV_TYPE) != 0) ntb->dev_type = NTB_DEV_USD; else ntb->dev_type = NTB_DEV_DSD; conn_type = ppd & XEON_PPD_CONN_TYPE; switch (conn_type) { case NTB_CONN_B2B: ntb->conn_type = conn_type; break; case NTB_CONN_RP: case NTB_CONN_TRANSPARENT: default: device_printf(ntb->device, "Unsupported connection type: %u\n", (unsigned)conn_type); return (ENXIO); } return (0); } static int ntb_detect_soc(struct ntb_softc *ntb) { uint32_t ppd, conn_type; ppd = pci_read_config(ntb->device, NTB_PPD_OFFSET, 4); ntb->ppd = ppd; if ((ppd & SOC_PPD_DEV_TYPE) != 0) ntb->dev_type = NTB_DEV_DSD; else ntb->dev_type = NTB_DEV_USD; conn_type = (ppd & SOC_PPD_CONN_TYPE) >> 8; switch (conn_type) { case NTB_CONN_B2B: ntb->conn_type = conn_type; break; default: device_printf(ntb->device, "Unsupported NTB configuration\n"); return (ENXIO); } return (0); } static int ntb_setup_xeon(struct ntb_softc *ntb) { ntb->reg_ofs.ldb = XEON_PDOORBELL_OFFSET; ntb->reg_ofs.ldb_mask = XEON_PDBMSK_OFFSET; ntb->reg_ofs.spad_local = XEON_SPAD_OFFSET; ntb->reg_ofs.bar2_xlat = XEON_SBAR2XLAT_OFFSET; ntb->reg_ofs.bar4_xlat = XEON_SBAR4XLAT_OFFSET; switch (ntb->conn_type) { case NTB_CONN_B2B: /* * reg_ofs.rdb and reg_ofs.spad_remote are effectively ignored * with the NTB_REGS_THRU_MW errata mode enabled. (See * ntb_ring_doorbell() and ntb_read/write_remote_spad().) */ ntb->reg_ofs.rdb = XEON_B2B_DOORBELL_OFFSET; ntb->reg_ofs.spad_remote = XEON_B2B_SPAD_OFFSET; ntb->limits.max_spads = XEON_MAX_SPADS; break; case NTB_CONN_RP: /* * Every Xeon today needs NTB_REGS_THRU_MW, so punt on RP for * now. */ KASSERT(HAS_FEATURE(NTB_REGS_THRU_MW), ("Xeon without MW errata unimplemented")); device_printf(ntb->device, "NTB-RP disabled to due hardware errata.\n"); return (ENXIO); case NTB_CONN_TRANSPARENT: default: device_printf(ntb->device, "Connection type %d not supported\n", ntb->conn_type); return (ENXIO); } /* * There is a Xeon hardware errata related to writes to SDOORBELL or * B2BDOORBELL in conjunction with inbound access to NTB MMIO space, * which may hang the system. To workaround this use the second memory * window to access the interrupt and scratch pad registers on the * remote system. * * There is another HW errata on the limit registers -- they can only * be written when the base register is (?)4GB aligned and < 32-bit. * This should already be the case based on the driver defaults, but * write the limit registers first just in case. */ - if (HAS_FEATURE(NTB_REGS_THRU_MW)) + if (HAS_FEATURE(NTB_REGS_THRU_MW)) { + /* Reserve the last MW for mapping remote spad */ + ntb->limits.max_mw--; /* * Set the Limit register to 4k, the minimum size, to prevent * an illegal access. */ ntb_reg_write(8, XEON_PBAR4LMT_OFFSET, ntb_get_mw_size(ntb, 1) + 0x1000); - else + } else /* * Disable the limit register, just in case it is set to * something silly. */ ntb_reg_write(8, XEON_PBAR4LMT_OFFSET, 0); ntb->reg_ofs.lnk_cntl = XEON_NTBCNTL_OFFSET; ntb->reg_ofs.lnk_stat = XEON_LINK_STATUS_OFFSET; ntb->reg_ofs.spci_cmd = XEON_PCICMD_OFFSET; ntb->limits.max_db_bits = XEON_MAX_DB_BITS; ntb->limits.msix_cnt = XEON_MSIX_CNT; ntb->bits_per_vector = XEON_DB_BITS_PER_VEC; /* * HW Errata on bit 14 of b2bdoorbell register. Writes will not be * mirrored to the remote system. Shrink the number of bits by one, * since bit 14 is the last bit. * * On REGS_THRU_MW errata mode, we don't use the b2bdoorbell register * anyway. Nor for non-B2B connection types. */ if (HAS_FEATURE(NTB_B2BDOORBELL_BIT14) && !HAS_FEATURE(NTB_REGS_THRU_MW) && ntb->conn_type == NTB_CONN_B2B) ntb->limits.max_db_bits = XEON_MAX_DB_BITS - 1; configure_xeon_secondary_side_bars(ntb); /* Enable Bus Master and Memory Space on the secondary side */ if (ntb->conn_type == NTB_CONN_B2B) ntb_reg_write(2, ntb->reg_ofs.spci_cmd, PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN); /* Enable link training */ ntb_hw_link_up(ntb); return (0); } static int ntb_setup_soc(struct ntb_softc *ntb) { KASSERT(ntb->conn_type == NTB_CONN_B2B, ("Unsupported NTB configuration (%d)\n", ntb->conn_type)); /* Initiate PCI-E link training */ pci_write_config(ntb->device, NTB_PPD_OFFSET, ntb->ppd | SOC_PPD_INIT_LINK, 4); ntb->reg_ofs.ldb = SOC_PDOORBELL_OFFSET; ntb->reg_ofs.ldb_mask = SOC_PDBMSK_OFFSET; ntb->reg_ofs.rdb = SOC_B2B_DOORBELL_OFFSET; ntb->reg_ofs.bar2_xlat = SOC_SBAR2XLAT_OFFSET; ntb->reg_ofs.bar4_xlat = SOC_SBAR4XLAT_OFFSET; ntb->reg_ofs.lnk_cntl = SOC_NTBCNTL_OFFSET; ntb->reg_ofs.lnk_stat = SOC_LINK_STATUS_OFFSET; ntb->reg_ofs.spad_local = SOC_SPAD_OFFSET; ntb->reg_ofs.spad_remote = SOC_B2B_SPAD_OFFSET; ntb->reg_ofs.spci_cmd = SOC_PCICMD_OFFSET; ntb->limits.max_spads = SOC_MAX_SPADS; ntb->limits.max_db_bits = SOC_MAX_DB_BITS; ntb->limits.msix_cnt = SOC_MSIX_CNT; ntb->bits_per_vector = SOC_DB_BITS_PER_VEC; /* * FIXME - MSI-X bug on early SOC HW, remove once internal issue is * resolved. Mask transaction layer internal parity errors. */ pci_write_config(ntb->device, 0xFC, 0x4, 4); configure_soc_secondary_side_bars(ntb); /* Enable Bus Master and Memory Space on the secondary side */ ntb_reg_write(2, ntb->reg_ofs.spci_cmd, PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN); callout_reset(&ntb->heartbeat_timer, 0, ntb_handle_heartbeat, ntb); return (0); } static void configure_soc_secondary_side_bars(struct ntb_softc *ntb) { if (ntb->dev_type == NTB_DEV_USD) { ntb_reg_write(8, SOC_PBAR2XLAT_OFFSET, PBAR2XLAT_USD_ADDR); ntb_reg_write(8, SOC_PBAR4XLAT_OFFSET, PBAR4XLAT_USD_ADDR); ntb_reg_write(8, SOC_MBAR23_OFFSET, MBAR23_USD_ADDR); ntb_reg_write(8, SOC_MBAR45_OFFSET, MBAR45_USD_ADDR); } else { ntb_reg_write(8, SOC_PBAR2XLAT_OFFSET, PBAR2XLAT_DSD_ADDR); ntb_reg_write(8, SOC_PBAR4XLAT_OFFSET, PBAR4XLAT_DSD_ADDR); ntb_reg_write(8, SOC_MBAR23_OFFSET, MBAR23_DSD_ADDR); ntb_reg_write(8, SOC_MBAR45_OFFSET, MBAR45_DSD_ADDR); } } static void configure_xeon_secondary_side_bars(struct ntb_softc *ntb) { if (ntb->dev_type == NTB_DEV_USD) { ntb_reg_write(8, XEON_PBAR2XLAT_OFFSET, PBAR2XLAT_USD_ADDR); if (HAS_FEATURE(NTB_REGS_THRU_MW)) ntb_reg_write(8, XEON_PBAR4XLAT_OFFSET, MBAR01_DSD_ADDR); else { ntb_reg_write(8, XEON_PBAR4XLAT_OFFSET, PBAR4XLAT_USD_ADDR); /* * B2B_XLAT_OFFSET is a 64-bit register but can only be * written 32 bits at a time. */ ntb_reg_write(4, XEON_B2B_XLAT_OFFSETL, MBAR01_DSD_ADDR & 0xffffffff); ntb_reg_write(4, XEON_B2B_XLAT_OFFSETU, MBAR01_DSD_ADDR >> 32); } ntb_reg_write(8, XEON_SBAR0BASE_OFFSET, MBAR01_USD_ADDR); ntb_reg_write(8, XEON_SBAR2BASE_OFFSET, MBAR23_USD_ADDR); ntb_reg_write(8, XEON_SBAR4BASE_OFFSET, MBAR45_USD_ADDR); } else { ntb_reg_write(8, XEON_PBAR2XLAT_OFFSET, PBAR2XLAT_DSD_ADDR); if (HAS_FEATURE(NTB_REGS_THRU_MW)) ntb_reg_write(8, XEON_PBAR4XLAT_OFFSET, MBAR01_USD_ADDR); else { ntb_reg_write(8, XEON_PBAR4XLAT_OFFSET, PBAR4XLAT_DSD_ADDR); /* * B2B_XLAT_OFFSET is a 64-bit register but can only be * written 32 bits at a time. */ ntb_reg_write(4, XEON_B2B_XLAT_OFFSETL, MBAR01_USD_ADDR & 0xffffffff); ntb_reg_write(4, XEON_B2B_XLAT_OFFSETU, MBAR01_USD_ADDR >> 32); } ntb_reg_write(8, XEON_SBAR0BASE_OFFSET, MBAR01_DSD_ADDR); ntb_reg_write(8, XEON_SBAR2BASE_OFFSET, MBAR23_DSD_ADDR); ntb_reg_write(8, XEON_SBAR4BASE_OFFSET, MBAR45_DSD_ADDR); } } /* SOC does not have link status interrupt, poll on that platform */ static void ntb_handle_heartbeat(void *arg) { struct ntb_softc *ntb = arg; uint32_t status32; int rc; rc = ntb_check_link_status(ntb); if (rc != 0) device_printf(ntb->device, "Error determining link status\n"); /* Check to see if a link error is the cause of the link down */ if (ntb->link_status == NTB_LINK_DOWN) { status32 = ntb_reg_read(4, SOC_LTSSMSTATEJMP_OFFSET); if ((status32 & SOC_LTSSMSTATEJMP_FORCEDETECT) != 0) { callout_reset(&ntb->lr_timer, 0, recover_soc_link, ntb); return; } } callout_reset(&ntb->heartbeat_timer, NTB_HB_TIMEOUT * hz, ntb_handle_heartbeat, ntb); } static void soc_perform_link_restart(struct ntb_softc *ntb) { uint32_t status; /* Driver resets the NTB ModPhy lanes - magic! */ ntb_reg_write(1, SOC_MODPHY_PCSREG6, 0xe0); ntb_reg_write(1, SOC_MODPHY_PCSREG4, 0x40); ntb_reg_write(1, SOC_MODPHY_PCSREG4, 0x60); ntb_reg_write(1, SOC_MODPHY_PCSREG6, 0x60); /* Driver waits 100ms to allow the NTB ModPhy to settle */ pause("ModPhy", hz / 10); /* Clear AER Errors, write to clear */ status = ntb_reg_read(4, SOC_ERRCORSTS_OFFSET); status &= PCIM_AER_COR_REPLAY_ROLLOVER; ntb_reg_write(4, SOC_ERRCORSTS_OFFSET, status); /* Clear unexpected electrical idle event in LTSSM, write to clear */ status = ntb_reg_read(4, SOC_LTSSMERRSTS0_OFFSET); status |= SOC_LTSSMERRSTS0_UNEXPECTEDEI; ntb_reg_write(4, SOC_LTSSMERRSTS0_OFFSET, status); /* Clear DeSkew Buffer error, write to clear */ status = ntb_reg_read(4, SOC_DESKEWSTS_OFFSET); status |= SOC_DESKEWSTS_DBERR; ntb_reg_write(4, SOC_DESKEWSTS_OFFSET, status); status = ntb_reg_read(4, SOC_IBSTERRRCRVSTS0_OFFSET); status &= SOC_IBIST_ERR_OFLOW; ntb_reg_write(4, SOC_IBSTERRRCRVSTS0_OFFSET, status); /* Releases the NTB state machine to allow the link to retrain */ status = ntb_reg_read(4, SOC_LTSSMSTATEJMP_OFFSET); status &= ~SOC_LTSSMSTATEJMP_FORCEDETECT; ntb_reg_write(4, SOC_LTSSMSTATEJMP_OFFSET, status); } static void ntb_handle_link_event(struct ntb_softc *ntb, int link_state) { enum ntb_hw_event event; uint16_t status; if (ntb->link_status == link_state) return; if (link_state == NTB_LINK_UP) { device_printf(ntb->device, "Link Up\n"); ntb->link_status = NTB_LINK_UP; event = NTB_EVENT_HW_LINK_UP; if (ntb->type == NTB_SOC || ntb->conn_type == NTB_CONN_TRANSPARENT) status = ntb_reg_read(2, ntb->reg_ofs.lnk_stat); else status = pci_read_config(ntb->device, XEON_LINK_STATUS_OFFSET, 2); ntb->link_width = (status & NTB_LINK_WIDTH_MASK) >> 4; ntb->link_speed = (status & NTB_LINK_SPEED_MASK); device_printf(ntb->device, "Link Width %d, Link Speed %d\n", ntb->link_width, ntb->link_speed); callout_reset(&ntb->heartbeat_timer, NTB_HB_TIMEOUT * hz, ntb_handle_heartbeat, ntb); } else { device_printf(ntb->device, "Link Down\n"); ntb->link_status = NTB_LINK_DOWN; event = NTB_EVENT_HW_LINK_DOWN; /* Do not modify link width/speed, we need it in link recovery */ } /* notify the upper layer if we have an event change */ if (ntb->event_cb != NULL) ntb->event_cb(ntb->ntb_transport, event); } static void ntb_hw_link_up(struct ntb_softc *ntb) { uint32_t cntl; if (ntb->conn_type == NTB_CONN_TRANSPARENT) { ntb_handle_link_event(ntb, NTB_LINK_UP); return; } cntl = ntb_reg_read(4, ntb->reg_ofs.lnk_cntl); cntl &= ~(NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK); cntl |= NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP; cntl |= NTB_CNTL_P2S_BAR45_SNOOP | NTB_CNTL_S2P_BAR45_SNOOP; ntb_reg_write(4, ntb->reg_ofs.lnk_cntl, cntl); } static void ntb_hw_link_down(struct ntb_softc *ntb) { uint32_t cntl; if (ntb->conn_type == NTB_CONN_TRANSPARENT) { ntb_handle_link_event(ntb, NTB_LINK_DOWN); return; } cntl = ntb_reg_read(4, ntb->reg_ofs.lnk_cntl); cntl &= ~(NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP); cntl &= ~(NTB_CNTL_P2S_BAR45_SNOOP | NTB_CNTL_S2P_BAR45_SNOOP); cntl |= NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK; ntb_reg_write(4, ntb->reg_ofs.lnk_cntl, cntl); } static void recover_soc_link(void *arg) { struct ntb_softc *ntb = arg; uint8_t speed, width; uint32_t status32; uint16_t status16; soc_perform_link_restart(ntb); /* * There is a potential race between the 2 NTB devices recovering at * the same time. If the times are the same, the link will not recover * and the driver will be stuck in this loop forever. Add a random * interval to the recovery time to prevent this race. */ status32 = arc4random() % SOC_LINK_RECOVERY_TIME; pause("Link", (SOC_LINK_RECOVERY_TIME + status32) * hz / 1000); status32 = ntb_reg_read(4, SOC_LTSSMSTATEJMP_OFFSET); if ((status32 & SOC_LTSSMSTATEJMP_FORCEDETECT) != 0) goto retry; status32 = ntb_reg_read(4, SOC_IBSTERRRCRVSTS0_OFFSET); if ((status32 & SOC_IBIST_ERR_OFLOW) != 0) goto retry; status32 = ntb_reg_read(4, ntb->reg_ofs.lnk_cntl); if ((status32 & SOC_CNTL_LINK_DOWN) != 0) goto out; status16 = ntb_reg_read(2, ntb->reg_ofs.lnk_stat); width = (status16 & NTB_LINK_WIDTH_MASK) >> 4; speed = (status16 & NTB_LINK_SPEED_MASK); if (ntb->link_width != width || ntb->link_speed != speed) goto retry; out: callout_reset(&ntb->heartbeat_timer, NTB_HB_TIMEOUT * hz, ntb_handle_heartbeat, ntb); return; retry: callout_reset(&ntb->lr_timer, NTB_HB_TIMEOUT * hz, recover_soc_link, ntb); } static int ntb_check_link_status(struct ntb_softc *ntb) { int link_state; uint32_t ntb_cntl; uint16_t status; if (ntb->type == NTB_SOC) { ntb_cntl = ntb_reg_read(4, ntb->reg_ofs.lnk_cntl); if ((ntb_cntl & SOC_CNTL_LINK_DOWN) != 0) link_state = NTB_LINK_DOWN; else link_state = NTB_LINK_UP; } else { status = pci_read_config(ntb->device, XEON_LINK_STATUS_OFFSET, 2); if ((status & NTB_LINK_STATUS_ACTIVE) != 0) link_state = NTB_LINK_UP; else link_state = NTB_LINK_DOWN; } ntb_handle_link_event(ntb, link_state); return (0); } /** * ntb_register_event_callback() - register event callback * @ntb: pointer to ntb_softc instance * @func: callback function to register * * This function registers a callback for any HW driver events such as link * up/down, power management notices and etc. * * RETURNS: An appropriate ERRNO error value on error, or zero for success. */ int ntb_register_event_callback(struct ntb_softc *ntb, ntb_event_callback func) { if (ntb->event_cb != NULL) return (EINVAL); ntb->event_cb = func; return (0); } /** * ntb_unregister_event_callback() - unregisters the event callback * @ntb: pointer to ntb_softc instance * * This function unregisters the existing callback from transport */ void ntb_unregister_event_callback(struct ntb_softc *ntb) { ntb->event_cb = NULL; } static void ntb_irq_work(void *arg) { struct ntb_db_cb *db_cb = arg; struct ntb_softc *ntb; int rc; rc = db_cb->callback(db_cb->data, db_cb->db_num); /* Poll if forward progress was made. */ if (rc != 0) { callout_reset(&db_cb->irq_work, 0, ntb_irq_work, db_cb); return; } /* Unmask interrupt if no progress was made. */ ntb = db_cb->ntb; unmask_ldb_interrupt(ntb, db_cb->db_num); } /** * ntb_register_db_callback() - register a callback for doorbell interrupt * @ntb: pointer to ntb_softc instance * @idx: doorbell index to register callback, zero based * @data: pointer to be returned to caller with every callback * @func: callback function to register * * This function registers a callback function for the doorbell interrupt * on the primary side. The function will unmask the doorbell as well to * allow interrupt. * * RETURNS: An appropriate ERRNO error value on error, or zero for success. */ int ntb_register_db_callback(struct ntb_softc *ntb, unsigned int idx, void *data, ntb_db_callback func) { struct ntb_db_cb *db_cb = &ntb->db_cb[idx]; - if (idx >= ntb->allocated_interrupts || db_cb->callback || - db_cb->reserved) { + if (idx >= ntb->max_cbs || db_cb->callback != NULL || db_cb->reserved) { device_printf(ntb->device, "Invalid Index.\n"); return (EINVAL); } db_cb->callback = func; db_cb->data = data; callout_init(&db_cb->irq_work, 1); unmask_ldb_interrupt(ntb, idx); return (0); } /** * ntb_unregister_db_callback() - unregister a callback for doorbell interrupt * @ntb: pointer to ntb_softc instance * @idx: doorbell index to register callback, zero based * * This function unregisters a callback function for the doorbell interrupt * on the primary side. The function will also mask the said doorbell. */ void ntb_unregister_db_callback(struct ntb_softc *ntb, unsigned int idx) { - if (idx >= ntb->allocated_interrupts || !ntb->db_cb[idx].callback) + if (idx >= ntb->max_cbs || ntb->db_cb[idx].callback == NULL) return; mask_ldb_interrupt(ntb, idx); callout_drain(&ntb->db_cb[idx].irq_work); ntb->db_cb[idx].callback = NULL; } /** * ntb_find_transport() - find the transport pointer * @transport: pointer to pci device * * Given the pci device pointer, return the transport pointer passed in when * the transport attached when it was inited. * * RETURNS: pointer to transport. */ void * ntb_find_transport(struct ntb_softc *ntb) { return (ntb->ntb_transport); } /** * ntb_register_transport() - Register NTB transport with NTB HW driver * @transport: transport identifier * * This function allows a transport to reserve the hardware driver for * NTB usage. * * RETURNS: pointer to ntb_softc, NULL on error. */ struct ntb_softc * ntb_register_transport(struct ntb_softc *ntb, void *transport) { /* * TODO: when we have more than one transport, we will need to rewrite * this to prevent race conditions */ if (ntb->ntb_transport != NULL) return (NULL); ntb->ntb_transport = transport; return (ntb); } /** * ntb_unregister_transport() - Unregister the transport with the NTB HW driver * @ntb - ntb_softc of the transport to be freed * * This function unregisters the transport from the HW driver and performs any * necessary cleanups. */ void ntb_unregister_transport(struct ntb_softc *ntb) { - int i; + uint8_t i; if (ntb->ntb_transport == NULL) return; - for (i = 0; i < ntb->allocated_interrupts; i++) + for (i = 0; i < ntb->max_cbs; i++) ntb_unregister_db_callback(ntb, i); ntb_unregister_event_callback(ntb); ntb->ntb_transport = NULL; } /** * ntb_get_max_spads() - get the total scratch regs usable * @ntb: pointer to ntb_softc instance * * This function returns the max 32bit scratchpad registers usable by the * upper layer. * * RETURNS: total number of scratch pad registers available */ uint8_t ntb_get_max_spads(struct ntb_softc *ntb) { return (ntb->limits.max_spads); } +uint8_t +ntb_get_max_cbs(struct ntb_softc *ntb) +{ + + return (ntb->max_cbs); +} + +uint8_t +ntb_get_max_mw(struct ntb_softc *ntb) +{ + + return (ntb->limits.max_mw); +} + /** * ntb_write_local_spad() - write to the secondary scratchpad register * @ntb: pointer to ntb_softc instance * @idx: index to the scratchpad register, 0 based * @val: the data value to put into the register * * This function allows writing of a 32bit value to the indexed scratchpad * register. The register resides on the secondary (external) side. * * RETURNS: An appropriate ERRNO error value on error, or zero for success. */ int ntb_write_local_spad(struct ntb_softc *ntb, unsigned int idx, uint32_t val) { if (idx >= ntb->limits.max_spads) return (EINVAL); ntb_reg_write(4, ntb->reg_ofs.spad_local + idx * 4, val); return (0); } /** * ntb_read_local_spad() - read from the primary scratchpad register * @ntb: pointer to ntb_softc instance * @idx: index to scratchpad register, 0 based * @val: pointer to 32bit integer for storing the register value * * This function allows reading of the 32bit scratchpad register on * the primary (internal) side. * * RETURNS: An appropriate ERRNO error value on error, or zero for success. */ int ntb_read_local_spad(struct ntb_softc *ntb, unsigned int idx, uint32_t *val) { if (idx >= ntb->limits.max_spads) return (EINVAL); *val = ntb_reg_read(4, ntb->reg_ofs.spad_local + idx * 4); return (0); } /** * ntb_write_remote_spad() - write to the secondary scratchpad register * @ntb: pointer to ntb_softc instance * @idx: index to the scratchpad register, 0 based * @val: the data value to put into the register * * This function allows writing of a 32bit value to the indexed scratchpad * register. The register resides on the secondary (external) side. * * RETURNS: An appropriate ERRNO error value on error, or zero for success. */ int ntb_write_remote_spad(struct ntb_softc *ntb, unsigned int idx, uint32_t val) { if (idx >= ntb->limits.max_spads) return (EINVAL); if (HAS_FEATURE(NTB_REGS_THRU_MW)) ntb_mw_write(4, XEON_SHADOW_SPAD_OFFSET + idx * 4, val); else ntb_reg_write(4, ntb->reg_ofs.spad_remote + idx * 4, val); return (0); } /** * ntb_read_remote_spad() - read from the primary scratchpad register * @ntb: pointer to ntb_softc instance * @idx: index to scratchpad register, 0 based * @val: pointer to 32bit integer for storing the register value * * This function allows reading of the 32bit scratchpad register on * the primary (internal) side. * * RETURNS: An appropriate ERRNO error value on error, or zero for success. */ int ntb_read_remote_spad(struct ntb_softc *ntb, unsigned int idx, uint32_t *val) { if (idx >= ntb->limits.max_spads) return (EINVAL); if (HAS_FEATURE(NTB_REGS_THRU_MW)) *val = ntb_mw_read(4, XEON_SHADOW_SPAD_OFFSET + idx * 4); else *val = ntb_reg_read(4, ntb->reg_ofs.spad_remote + idx * 4); return (0); } /** * ntb_get_mw_vbase() - get virtual addr for the NTB memory window * @ntb: pointer to ntb_softc instance * @mw: memory window number * * This function provides the base virtual address of the memory window * specified. * * RETURNS: pointer to virtual address, or NULL on error. */ void * ntb_get_mw_vbase(struct ntb_softc *ntb, unsigned int mw) { - if (mw >= NTB_NUM_MW) + if (mw >= ntb_get_max_mw(ntb)) return (NULL); return (ntb->bar_info[NTB_MW_TO_BAR(mw)].vbase); } vm_paddr_t ntb_get_mw_pbase(struct ntb_softc *ntb, unsigned int mw) { - if (mw >= NTB_NUM_MW) + if (mw >= ntb_get_max_mw(ntb)) return (0); return (ntb->bar_info[NTB_MW_TO_BAR(mw)].pbase); } /** * ntb_get_mw_size() - return size of NTB memory window * @ntb: pointer to ntb_softc instance * @mw: memory window number * * This function provides the physical size of the memory window specified * * RETURNS: the size of the memory window or zero on error */ u_long ntb_get_mw_size(struct ntb_softc *ntb, unsigned int mw) { - if (mw >= NTB_NUM_MW) + if (mw >= ntb_get_max_mw(ntb)) return (0); return (ntb->bar_info[NTB_MW_TO_BAR(mw)].size); } /** * ntb_set_mw_addr - set the memory window address * @ntb: pointer to ntb_softc instance * @mw: memory window number * @addr: base address for data * * This function sets the base physical address of the memory window. This * memory address is where data from the remote system will be transfered into * or out of depending on how the transport is configured. */ void ntb_set_mw_addr(struct ntb_softc *ntb, unsigned int mw, uint64_t addr) { - if (mw >= NTB_NUM_MW) + if (mw >= ntb_get_max_mw(ntb)) return; switch (NTB_MW_TO_BAR(mw)) { case NTB_B2B_BAR_1: ntb_reg_write(8, ntb->reg_ofs.bar2_xlat, addr); break; case NTB_B2B_BAR_2: ntb_reg_write(8, ntb->reg_ofs.bar4_xlat, addr); break; } } /** * ntb_ring_doorbell() - Set the doorbell on the secondary/external side * @ntb: pointer to ntb_softc instance * @db: doorbell to ring * * This function allows triggering of a doorbell on the secondary/external * side that will initiate an interrupt on the remote host */ void ntb_ring_doorbell(struct ntb_softc *ntb, unsigned int db) { uint64_t bit; if (ntb->type == NTB_SOC) bit = 1 << db; else bit = ((1 << ntb->bits_per_vector) - 1) << (db * ntb->bits_per_vector); if (HAS_FEATURE(NTB_REGS_THRU_MW)) { ntb_mw_write(2, XEON_SHADOW_PDOORBELL_OFFSET, bit); return; } db_iowrite(ntb, ntb->reg_ofs.rdb, bit); } /** * ntb_query_link_status() - return the hardware link status * @ndev: pointer to ntb_device instance * * Returns true if the hardware is connected to the remote system * * RETURNS: true or false based on the hardware link state */ bool ntb_query_link_status(struct ntb_softc *ntb) { return (ntb->link_status == NTB_LINK_UP); } static void save_bar_parameters(struct ntb_pci_bar_info *bar) { bar->pci_bus_tag = rman_get_bustag(bar->pci_resource); bar->pci_bus_handle = rman_get_bushandle(bar->pci_resource); bar->pbase = rman_get_start(bar->pci_resource); bar->size = rman_get_size(bar->pci_resource); bar->vbase = rman_get_virtual(bar->pci_resource); } device_t ntb_get_device(struct ntb_softc *ntb) { return (ntb->device); } /* Export HW-specific errata information. */ bool ntb_has_feature(struct ntb_softc *ntb, uint64_t feature) { return (HAS_FEATURE(feature)); } diff --git a/sys/dev/ntb/ntb_hw/ntb_hw.h b/sys/dev/ntb/ntb_hw/ntb_hw.h index 7abd10c54dbf..b6b85a02448e 100644 --- a/sys/dev/ntb/ntb_hw/ntb_hw.h +++ b/sys/dev/ntb/ntb_hw/ntb_hw.h @@ -1,86 +1,88 @@ /*- * Copyright (C) 2013 Intel Corporation * All rights reserved. * * 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. * * $FreeBSD$ */ #ifndef _NTB_HW_H_ #define _NTB_HW_H_ struct ntb_softc; -#define NTB_NUM_MW 2 +#define NTB_MAX_NUM_MW 2 enum ntb_link_event { NTB_LINK_DOWN = 0, NTB_LINK_UP, }; enum ntb_hw_event { NTB_EVENT_SW_EVENT0 = 0, NTB_EVENT_SW_EVENT1, NTB_EVENT_SW_EVENT2, NTB_EVENT_HW_ERROR, NTB_EVENT_HW_LINK_UP, NTB_EVENT_HW_LINK_DOWN, }; SYSCTL_DECL(_hw_ntb); typedef int (*ntb_db_callback)(void *data, int db_num); typedef void (*ntb_event_callback)(void *data, enum ntb_hw_event event); int ntb_register_event_callback(struct ntb_softc *ntb, ntb_event_callback func); void ntb_unregister_event_callback(struct ntb_softc *ntb); int ntb_register_db_callback(struct ntb_softc *ntb, unsigned int idx, void *data, ntb_db_callback func); void ntb_unregister_db_callback(struct ntb_softc *ntb, unsigned int idx); void *ntb_find_transport(struct ntb_softc *ntb); struct ntb_softc *ntb_register_transport(struct ntb_softc *ntb, void *transport); void ntb_unregister_transport(struct ntb_softc *ntb); +uint8_t ntb_get_max_cbs(struct ntb_softc *ntb); +uint8_t ntb_get_max_mw(struct ntb_softc *ntb); uint8_t ntb_get_max_spads(struct ntb_softc *ntb); int ntb_write_local_spad(struct ntb_softc *ntb, unsigned int idx, uint32_t val); int ntb_read_local_spad(struct ntb_softc *ntb, unsigned int idx, uint32_t *val); int ntb_write_remote_spad(struct ntb_softc *ntb, unsigned int idx, uint32_t val); int ntb_read_remote_spad(struct ntb_softc *ntb, unsigned int idx, uint32_t *val); void *ntb_get_mw_vbase(struct ntb_softc *ntb, unsigned int mw); vm_paddr_t ntb_get_mw_pbase(struct ntb_softc *ntb, unsigned int mw); u_long ntb_get_mw_size(struct ntb_softc *ntb, unsigned int mw); void ntb_set_mw_addr(struct ntb_softc *ntb, unsigned int mw, uint64_t addr); void ntb_ring_doorbell(struct ntb_softc *ntb, unsigned int db); bool ntb_query_link_status(struct ntb_softc *ntb); device_t ntb_get_device(struct ntb_softc *ntb); #define NTB_BAR_SIZE_4K (1 << 0) /* REGS_THRU_MW is the equivalent of Linux's NTB_HWERR_SDOORBELL_LOCKUP */ #define NTB_REGS_THRU_MW (1 << 1) #define NTB_SB01BASE_LOCKUP (1 << 2) #define NTB_B2BDOORBELL_BIT14 (1 << 3) bool ntb_has_feature(struct ntb_softc *, uint64_t); #endif /* _NTB_HW_H_ */