diff --git a/sys/dev/xen/netfront/netfront.c b/sys/dev/xen/netfront/netfront.c index cef6370493c2..e940d93bfe05 100644 --- a/sys/dev/xen/netfront/netfront.c +++ b/sys/dev/xen/netfront/netfront.c @@ -1,5 +1,6 @@ /*- * Copyright (c) 2004-2006 Kip Macy + * Copyright (c) 2015 Wei Liu * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -39,15 +40,14 @@ __FBSDID("$FreeBSD$"); #include #include #include +#include #include #include #include #include #include - #include - #include #include @@ -86,6 +86,12 @@ __FBSDID("$FreeBSD$"); static int xn_enable_lro = 1; TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro); +/* + * Number of pairs of queues. + */ +static unsigned long xn_num_queues = 4; +TUNABLE_ULONG("hw.xn.num_queues", &xn_num_queues); + /** * \brief The maximum allowed data fragments in a single transmit * request. @@ -100,149 +106,167 @@ TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro); #define net_ratelimit() 0 +struct netfront_rxq; +struct netfront_txq; struct netfront_info; struct netfront_rx_info; -static void xn_txeof(struct netfront_info *); -static void xn_rxeof(struct netfront_info *); -static void network_alloc_rx_buffers(struct netfront_info *); +static void xn_txeof(struct netfront_txq *); +static void xn_rxeof(struct netfront_rxq *); +static void xn_alloc_rx_buffers(struct netfront_rxq *); -static void xn_tick_locked(struct netfront_info *); -static void xn_tick(void *); +static void xn_release_rx_bufs(struct netfront_rxq *); +static void xn_release_tx_bufs(struct netfront_txq *); -static void xn_intr(void *); +static void xn_rxq_intr(void *); +static void xn_txq_intr(void *); +static int xn_intr(void *); static inline int xn_count_frags(struct mbuf *m); -static int xn_assemble_tx_request(struct netfront_info *sc, - struct mbuf *m_head); -static void xn_start_locked(struct ifnet *); -static void xn_start(struct ifnet *); -static int xn_ioctl(struct ifnet *, u_long, caddr_t); +static int xn_assemble_tx_request(struct netfront_txq *, struct mbuf *); +static int xn_ioctl(struct ifnet *, u_long, caddr_t); static void xn_ifinit_locked(struct netfront_info *); static void xn_ifinit(void *); static void xn_stop(struct netfront_info *); static void xn_query_features(struct netfront_info *np); -static int xn_configure_features(struct netfront_info *np); -#ifdef notyet -static void xn_watchdog(struct ifnet *); -#endif - -#ifdef notyet -static void netfront_closing(device_t dev); -#endif +static int xn_configure_features(struct netfront_info *np); static void netif_free(struct netfront_info *info); static int netfront_detach(device_t dev); +static int xn_txq_mq_start_locked(struct netfront_txq *, struct mbuf *); +static int xn_txq_mq_start(struct ifnet *, struct mbuf *); + static int talk_to_backend(device_t dev, struct netfront_info *info); static int create_netdev(device_t dev); static void netif_disconnect_backend(struct netfront_info *info); -static int setup_device(device_t dev, struct netfront_info *info); -static void free_ring(int *ref, void *ring_ptr_ref); - -static int xn_ifmedia_upd(struct ifnet *ifp); +static int setup_device(device_t dev, struct netfront_info *info, + unsigned long); +static int xn_ifmedia_upd(struct ifnet *ifp); static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr); -/* Xenolinux helper functions */ -int network_connect(struct netfront_info *); +int xn_connect(struct netfront_info *); -static void xn_free_rx_ring(struct netfront_info *); - -static void xn_free_tx_ring(struct netfront_info *); - -static int xennet_get_responses(struct netfront_info *np, - struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons, - struct mbuf **list); +static int xn_get_responses(struct netfront_rxq *, + struct netfront_rx_info *, RING_IDX, RING_IDX *, + struct mbuf **); #define virt_to_mfn(x) (vtophys(x) >> PAGE_SHIFT) #define INVALID_P2M_ENTRY (~0UL) -/* - * Mbuf pointers. We need these to keep track of the virtual addresses - * of our mbuf chains since we can only convert from virtual to physical, - * not the other way around. The size must track the free index arrays. - */ -struct xn_chain_data { - struct mbuf *xn_tx_chain[NET_TX_RING_SIZE+1]; - int xn_tx_chain_cnt; - struct mbuf *xn_rx_chain[NET_RX_RING_SIZE+1]; +struct xn_rx_stats +{ + u_long rx_packets; /* total packets received */ + u_long rx_bytes; /* total bytes received */ + u_long rx_errors; /* bad packets received */ }; -struct netfront_stats +struct xn_tx_stats { - u_long rx_packets; /* total packets received */ - u_long tx_packets; /* total packets transmitted */ - u_long rx_bytes; /* total bytes received */ - u_long tx_bytes; /* total bytes transmitted */ - u_long rx_errors; /* bad packets received */ - u_long tx_errors; /* packet transmit problems */ + u_long tx_packets; /* total packets transmitted */ + u_long tx_bytes; /* total bytes transmitted */ + u_long tx_errors; /* packet transmit problems */ +}; + +#define XN_QUEUE_NAME_LEN 8 /* xn{t,r}x_%u, allow for two digits */ +struct netfront_rxq { + struct netfront_info *info; + u_int id; + char name[XN_QUEUE_NAME_LEN]; + struct mtx lock; + + int ring_ref; + netif_rx_front_ring_t ring; + xen_intr_handle_t xen_intr_handle; + + grant_ref_t gref_head; + grant_ref_t grant_ref[NET_TX_RING_SIZE + 1]; + + struct mbuf *mbufs[NET_RX_RING_SIZE + 1]; + struct mbufq batch; /* batch queue */ + int target; + + xen_pfn_t pfn_array[NET_RX_RING_SIZE]; + + struct lro_ctrl lro; + + struct taskqueue *tq; + struct task intrtask; + + struct xn_rx_stats stats; +}; + +struct netfront_txq { + struct netfront_info *info; + u_int id; + char name[XN_QUEUE_NAME_LEN]; + struct mtx lock; + + int ring_ref; + netif_tx_front_ring_t ring; + xen_intr_handle_t xen_intr_handle; + + grant_ref_t gref_head; + grant_ref_t grant_ref[NET_TX_RING_SIZE + 1]; + + struct mbuf *mbufs[NET_TX_RING_SIZE + 1]; + int mbufs_cnt; + struct buf_ring *br; + + struct taskqueue *tq; + struct task intrtask; + struct task defrtask; + + bool full; + + struct xn_tx_stats stats; }; struct netfront_info { - struct ifnet *xn_ifp; - struct lro_ctrl xn_lro; + struct ifnet *xn_ifp; - struct netfront_stats stats; - u_int tx_full; + struct mtx sc_lock; - netif_tx_front_ring_t tx; - netif_rx_front_ring_t rx; + u_int num_queues; + struct netfront_rxq *rxq; + struct netfront_txq *txq; - struct mtx tx_lock; - struct mtx rx_lock; - struct mtx sc_lock; - - xen_intr_handle_t xen_intr_handle; - u_int carrier; - u_int maxfrags; + u_int carrier; + u_int maxfrags; /* Receive-ring batched refills. */ #define RX_MIN_TARGET 32 #define RX_MAX_TARGET NET_RX_RING_SIZE - int rx_min_target; - int rx_max_target; - int rx_target; - - grant_ref_t gref_tx_head; - grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1]; - grant_ref_t gref_rx_head; - grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1]; + int rx_min_target; + int rx_max_target; device_t xbdev; - int tx_ring_ref; - int rx_ring_ref; uint8_t mac[ETHER_ADDR_LEN]; - struct xn_chain_data xn_cdata; /* mbufs */ - struct mbufq xn_rx_batch; /* batch queue */ int xn_if_flags; - struct callout xn_stat_ch; - xen_pfn_t rx_pfn_array[NET_RX_RING_SIZE]; struct ifmedia sc_media; bool xn_resume; }; -#define rx_mbufs xn_cdata.xn_rx_chain -#define tx_mbufs xn_cdata.xn_tx_chain +struct netfront_rx_info { + struct netif_rx_response rx; + struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; +}; -#define XN_RX_LOCK(_sc) mtx_lock(&(_sc)->rx_lock) -#define XN_RX_UNLOCK(_sc) mtx_unlock(&(_sc)->rx_lock) +#define XN_RX_LOCK(_q) mtx_lock(&(_q)->lock) +#define XN_RX_UNLOCK(_q) mtx_unlock(&(_q)->lock) -#define XN_TX_LOCK(_sc) mtx_lock(&(_sc)->tx_lock) -#define XN_TX_UNLOCK(_sc) mtx_unlock(&(_sc)->tx_lock) +#define XN_TX_LOCK(_q) mtx_lock(&(_q)->lock) +#define XN_TX_TRYLOCK(_q) mtx_trylock(&(_q)->lock) +#define XN_TX_UNLOCK(_q) mtx_unlock(&(_q)->lock) #define XN_LOCK(_sc) mtx_lock(&(_sc)->sc_lock); #define XN_UNLOCK(_sc) mtx_unlock(&(_sc)->sc_lock); #define XN_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->sc_lock, MA_OWNED); -#define XN_RX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->rx_lock, MA_OWNED); -#define XN_TX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->tx_lock, MA_OWNED); - -struct netfront_rx_info { - struct netif_rx_response rx; - struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; -}; +#define XN_RX_LOCK_ASSERT(_q) mtx_assert(&(_q)->lock, MA_OWNED); +#define XN_TX_LOCK_ASSERT(_q) mtx_assert(&(_q)->lock, MA_OWNED); #define netfront_carrier_on(netif) ((netif)->carrier = 1) #define netfront_carrier_off(netif) ((netif)->carrier = 0) @@ -253,6 +277,7 @@ struct netfront_rx_info { static inline void add_id_to_freelist(struct mbuf **list, uintptr_t id) { + KASSERT(id != 0, ("%s: the head item (0) must always be free.", __func__)); list[id] = list[0]; @@ -272,30 +297,33 @@ get_id_from_freelist(struct mbuf **list) } static inline int -xennet_rxidx(RING_IDX idx) +xn_rxidx(RING_IDX idx) { + return idx & (NET_RX_RING_SIZE - 1); } static inline struct mbuf * -xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri) +xn_get_rx_mbuf(struct netfront_rxq *rxq, RING_IDX ri) { - int i = xennet_rxidx(ri); + int i; struct mbuf *m; - m = np->rx_mbufs[i]; - np->rx_mbufs[i] = NULL; + i = xn_rxidx(ri); + m = rxq->mbufs[i]; + rxq->mbufs[i] = NULL; return (m); } static inline grant_ref_t -xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri) +xn_get_rx_ref(struct netfront_rxq *rxq, RING_IDX ri) { - int i = xennet_rxidx(ri); - grant_ref_t ref = np->grant_rx_ref[i]; + int i = xn_rxidx(ri); + grant_ref_t ref = rxq->grant_ref[i]; + KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n")); - np->grant_rx_ref[i] = GRANT_REF_INVALID; - return ref; + rxq->grant_ref[i] = GRANT_REF_INVALID; + return (ref); } #define IPRINTK(fmt, args...) \ @@ -392,7 +420,7 @@ netfront_attach(device_t dev) int err; err = create_netdev(dev); - if (err) { + if (err != 0) { xenbus_dev_fatal(dev, err, "creating netdev"); return (err); } @@ -402,19 +430,29 @@ netfront_attach(device_t dev) OID_AUTO, "enable_lro", CTLFLAG_RW, &xn_enable_lro, 0, "Large Receive Offload"); + SYSCTL_ADD_ULONG(device_get_sysctl_ctx(dev), + SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), + OID_AUTO, "num_queues", CTLFLAG_RD, + &xn_num_queues, "Number of pairs of queues"); + return (0); } static int netfront_suspend(device_t dev) { - struct netfront_info *info = device_get_softc(dev); + struct netfront_info *np = device_get_softc(dev); + u_int i; - XN_RX_LOCK(info); - XN_TX_LOCK(info); - netfront_carrier_off(info); - XN_TX_UNLOCK(info); - XN_RX_UNLOCK(info); + for (i = 0; i < np->num_queues; i++) { + XN_RX_LOCK(&np->rxq[i]); + XN_TX_LOCK(&np->txq[i]); + } + netfront_carrier_off(np); + for (i = 0; i < np->num_queues; i++) { + XN_RX_UNLOCK(&np->rxq[i]); + XN_TX_UNLOCK(&np->txq[i]); + } return (0); } @@ -434,6 +472,61 @@ netfront_resume(device_t dev) return (0); } +static int +write_queue_xenstore_keys(device_t dev, + struct netfront_rxq *rxq, + struct netfront_txq *txq, + struct xs_transaction *xst, bool hierarchy) +{ + int err; + const char *message; + const char *node = xenbus_get_node(dev); + char *path; + size_t path_size; + + KASSERT(rxq->id == txq->id, ("Mismatch between RX and TX queue ids")); + /* Split event channel support is not yet there. */ + KASSERT(rxq->xen_intr_handle == txq->xen_intr_handle, + ("Split event channels are not supported")); + + if (hierarchy) { + path_size = strlen(node) + 10; + path = malloc(path_size, M_DEVBUF, M_WAITOK|M_ZERO); + snprintf(path, path_size, "%s/queue-%u", node, rxq->id); + } else { + path_size = strlen(node) + 1; + path = malloc(path_size, M_DEVBUF, M_WAITOK|M_ZERO); + snprintf(path, path_size, "%s", node); + } + + err = xs_printf(*xst, path, "tx-ring-ref","%u", txq->ring_ref); + if (err != 0) { + message = "writing tx ring-ref"; + goto error; + } + err = xs_printf(*xst, path, "rx-ring-ref","%u", rxq->ring_ref); + if (err != 0) { + message = "writing rx ring-ref"; + goto error; + } + err = xs_printf(*xst, path, "event-channel", "%u", + xen_intr_port(rxq->xen_intr_handle)); + if (err != 0) { + message = "writing event-channel"; + goto error; + } + + free(path, M_DEVBUF); + + return (0); + +error: + free(path, M_DEVBUF); + xenbus_dev_fatal(dev, err, "%s", message); + + return (err); +} + /* Common code used when first setting up, and when resuming. */ static int talk_to_backend(device_t dev, struct netfront_info *info) @@ -442,134 +535,427 @@ talk_to_backend(device_t dev, struct netfront_info *info) struct xs_transaction xst; const char *node = xenbus_get_node(dev); int err; + unsigned long num_queues, max_queues = 0; + unsigned int i; err = xen_net_read_mac(dev, info->mac); - if (err) { + if (err != 0) { xenbus_dev_fatal(dev, err, "parsing %s/mac", node); goto out; } - /* Create shared ring, alloc event channel. */ - err = setup_device(dev, info); - if (err) + err = xs_scanf(XST_NIL, xenbus_get_otherend_path(info->xbdev), + "multi-queue-max-queues", NULL, "%lu", &max_queues); + if (err != 0) + max_queues = 1; + num_queues = xn_num_queues; + if (num_queues > max_queues) + num_queues = max_queues; + + err = setup_device(dev, info, num_queues); + if (err != 0) goto out; again: err = xs_transaction_start(&xst); - if (err) { + if (err != 0) { xenbus_dev_fatal(dev, err, "starting transaction"); - goto destroy_ring; + goto free; } - err = xs_printf(xst, node, "tx-ring-ref","%u", - info->tx_ring_ref); - if (err) { - message = "writing tx ring-ref"; - goto abort_transaction; - } - err = xs_printf(xst, node, "rx-ring-ref","%u", - info->rx_ring_ref); - if (err) { - message = "writing rx ring-ref"; - goto abort_transaction; - } - err = xs_printf(xst, node, - "event-channel", "%u", - xen_intr_port(info->xen_intr_handle)); - if (err) { - message = "writing event-channel"; - goto abort_transaction; + + if (info->num_queues == 1) { + err = write_queue_xenstore_keys(dev, &info->rxq[0], + &info->txq[0], &xst, false); + if (err != 0) + goto abort_transaction_no_def_error; + } else { + err = xs_printf(xst, node, "multi-queue-num-queues", + "%u", info->num_queues); + if (err != 0) { + message = "writing multi-queue-num-queues"; + goto abort_transaction; + } + + for (i = 0; i < info->num_queues; i++) { + err = write_queue_xenstore_keys(dev, &info->rxq[i], + &info->txq[i], &xst, true); + if (err != 0) + goto abort_transaction_no_def_error; + } } + err = xs_printf(xst, node, "request-rx-copy", "%u", 1); - if (err) { + if (err != 0) { message = "writing request-rx-copy"; goto abort_transaction; } err = xs_printf(xst, node, "feature-rx-notify", "%d", 1); - if (err) { + if (err != 0) { message = "writing feature-rx-notify"; goto abort_transaction; } err = xs_printf(xst, node, "feature-sg", "%d", 1); - if (err) { + if (err != 0) { message = "writing feature-sg"; goto abort_transaction; } err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1); - if (err) { + if (err != 0) { message = "writing feature-gso-tcpv4"; goto abort_transaction; } err = xs_transaction_end(xst, 0); - if (err) { + if (err != 0) { if (err == EAGAIN) goto again; xenbus_dev_fatal(dev, err, "completing transaction"); - goto destroy_ring; + goto free; } return 0; abort_transaction: - xs_transaction_end(xst, 1); xenbus_dev_fatal(dev, err, "%s", message); - destroy_ring: + abort_transaction_no_def_error: + xs_transaction_end(xst, 1); + free: netif_free(info); out: - return err; + return (err); +} + +static void +xn_rxq_tq_intr(void *xrxq, int pending) +{ + struct netfront_rxq *rxq = xrxq; + + XN_RX_LOCK(rxq); + xn_rxeof(rxq); + XN_RX_UNLOCK(rxq); +} + +static void +xn_txq_start(struct netfront_txq *txq) +{ + struct netfront_info *np = txq->info; + struct ifnet *ifp = np->xn_ifp; + + XN_TX_LOCK_ASSERT(txq); + if (!drbr_empty(ifp, txq->br)) + xn_txq_mq_start_locked(txq, NULL); +} + +static void +xn_txq_tq_intr(void *xtxq, int pending) +{ + struct netfront_txq *txq = xtxq; + + XN_TX_LOCK(txq); + if (RING_HAS_UNCONSUMED_RESPONSES(&txq->ring)) + xn_txeof(txq); + xn_txq_start(txq); + XN_TX_UNLOCK(txq); +} + +static void +xn_txq_tq_deferred(void *xtxq, int pending) +{ + struct netfront_txq *txq = xtxq; + + XN_TX_LOCK(txq); + xn_txq_start(txq); + XN_TX_UNLOCK(txq); +} + +static void +disconnect_rxq(struct netfront_rxq *rxq) +{ + + xn_release_rx_bufs(rxq); + gnttab_free_grant_references(rxq->gref_head); + gnttab_end_foreign_access_ref(rxq->ring_ref); + /* + * No split event channel support at the moment, handle will + * be unbound in tx. So no need to call xen_intr_unbind here, + * but we do want to reset the handler to 0. + */ + rxq->xen_intr_handle = 0; +} + +static void +destroy_rxq(struct netfront_rxq *rxq) +{ + + free(rxq->ring.sring, M_DEVBUF); + taskqueue_drain_all(rxq->tq); + taskqueue_free(rxq->tq); +} + +static void +destroy_rxqs(struct netfront_info *np) +{ + int i; + + for (i = 0; i < np->num_queues; i++) + destroy_rxq(&np->rxq[i]); + + free(np->rxq, M_DEVBUF); + np->rxq = NULL; } static int -setup_device(device_t dev, struct netfront_info *info) +setup_rxqs(device_t dev, struct netfront_info *info, + unsigned long num_queues) { - netif_tx_sring_t *txs; - netif_rx_sring_t *rxs; + int q, i; int error; + netif_rx_sring_t *rxs; + struct netfront_rxq *rxq; - info->tx_ring_ref = GRANT_REF_INVALID; - info->rx_ring_ref = GRANT_REF_INVALID; - info->rx.sring = NULL; - info->tx.sring = NULL; + info->rxq = malloc(sizeof(struct netfront_rxq) * num_queues, + M_DEVBUF, M_WAITOK|M_ZERO); - txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO); - if (!txs) { - error = ENOMEM; - xenbus_dev_fatal(dev, error, "allocating tx ring page"); - goto fail; - } - SHARED_RING_INIT(txs); - FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE); - error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref); - if (error) - goto fail; + for (q = 0; q < num_queues; q++) { + rxq = &info->rxq[q]; - rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO); - if (!rxs) { - error = ENOMEM; - xenbus_dev_fatal(dev, error, "allocating rx ring page"); - goto fail; - } - SHARED_RING_INIT(rxs); - FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE); + rxq->id = q; + rxq->info = info; + rxq->target = RX_MIN_TARGET; + rxq->ring_ref = GRANT_REF_INVALID; + rxq->ring.sring = NULL; + snprintf(rxq->name, XN_QUEUE_NAME_LEN, "xnrx_%u", q); + mtx_init(&rxq->lock, rxq->name, "netfront receive lock", + MTX_DEF); - error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref); - if (error) - goto fail; + for (i = 0; i <= NET_RX_RING_SIZE; i++) { + rxq->mbufs[i] = NULL; + rxq->grant_ref[i] = GRANT_REF_INVALID; + } - error = xen_intr_alloc_and_bind_local_port(dev, - xenbus_get_otherend_id(dev), /*filter*/NULL, xn_intr, info, - INTR_TYPE_NET | INTR_MPSAFE | INTR_ENTROPY, &info->xen_intr_handle); + mbufq_init(&rxq->batch, INT_MAX); - if (error) { - xenbus_dev_fatal(dev, error, - "xen_intr_alloc_and_bind_local_port failed"); - goto fail; + /* Start resources allocation */ + + if (gnttab_alloc_grant_references(RX_MAX_TARGET, + &rxq->gref_head) != 0) { + device_printf(dev, "allocating rx gref"); + error = ENOMEM; + goto fail; + } + + rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, + M_WAITOK|M_ZERO); + SHARED_RING_INIT(rxs); + FRONT_RING_INIT(&rxq->ring, rxs, PAGE_SIZE); + + error = xenbus_grant_ring(dev, virt_to_mfn(rxs), + &rxq->ring_ref); + if (error != 0) { + device_printf(dev, "granting rx ring page"); + goto fail_grant_ring; + } + + TASK_INIT(&rxq->intrtask, 0, xn_rxq_tq_intr, rxq); + rxq->tq = taskqueue_create_fast(rxq->name, M_WAITOK, + taskqueue_thread_enqueue, &rxq->tq); + + error = taskqueue_start_threads(&rxq->tq, 1, PI_NET, + "%s rxq %d", device_get_nameunit(dev), rxq->id); + if (error != 0) { + device_printf(dev, "failed to start rx taskq %d\n", + rxq->id); + goto fail_start_thread; + } } return (0); - fail: - netif_free(info); +fail_start_thread: + gnttab_end_foreign_access_ref(rxq->ring_ref); + taskqueue_drain_all(rxq->tq); + taskqueue_free(rxq->tq); +fail_grant_ring: + gnttab_free_grant_references(rxq->gref_head); + free(rxq->ring.sring, M_DEVBUF); +fail: + for (; q >= 0; q--) { + disconnect_rxq(&info->rxq[q]); + destroy_rxq(&info->rxq[q]); + } + + free(info->rxq, M_DEVBUF); + return (error); +} + +static void +disconnect_txq(struct netfront_txq *txq) +{ + + xn_release_tx_bufs(txq); + gnttab_free_grant_references(txq->gref_head); + gnttab_end_foreign_access_ref(txq->ring_ref); + xen_intr_unbind(&txq->xen_intr_handle); +} + +static void +destroy_txq(struct netfront_txq *txq) +{ + + free(txq->ring.sring, M_DEVBUF); + buf_ring_free(txq->br, M_DEVBUF); + taskqueue_drain_all(txq->tq); + taskqueue_free(txq->tq); +} + +static void +destroy_txqs(struct netfront_info *np) +{ + int i; + + for (i = 0; i < np->num_queues; i++) + destroy_txq(&np->txq[i]); + + free(np->txq, M_DEVBUF); + np->txq = NULL; +} + +static int +setup_txqs(device_t dev, struct netfront_info *info, + unsigned long num_queues) +{ + int q, i; + int error; + netif_tx_sring_t *txs; + struct netfront_txq *txq; + + info->txq = malloc(sizeof(struct netfront_txq) * num_queues, + M_DEVBUF, M_WAITOK|M_ZERO); + + for (q = 0; q < num_queues; q++) { + txq = &info->txq[q]; + + txq->id = q; + txq->info = info; + + txq->ring_ref = GRANT_REF_INVALID; + txq->ring.sring = NULL; + + snprintf(txq->name, XN_QUEUE_NAME_LEN, "xntx_%u", q); + + mtx_init(&txq->lock, txq->name, "netfront transmit lock", + MTX_DEF); + + for (i = 0; i <= NET_TX_RING_SIZE; i++) { + txq->mbufs[i] = (void *) ((u_long) i+1); + txq->grant_ref[i] = GRANT_REF_INVALID; + } + txq->mbufs[NET_TX_RING_SIZE] = (void *)0; + + /* Start resources allocation. */ + + if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, + &txq->gref_head) != 0) { + device_printf(dev, "failed to allocate tx grant refs\n"); + error = ENOMEM; + goto fail; + } + + txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, + M_WAITOK|M_ZERO); + SHARED_RING_INIT(txs); + FRONT_RING_INIT(&txq->ring, txs, PAGE_SIZE); + + error = xenbus_grant_ring(dev, virt_to_mfn(txs), + &txq->ring_ref); + if (error != 0) { + device_printf(dev, "failed to grant tx ring\n"); + goto fail_grant_ring; + } + + txq->br = buf_ring_alloc(NET_TX_RING_SIZE, M_DEVBUF, + M_WAITOK, &txq->lock); + TASK_INIT(&txq->defrtask, 0, xn_txq_tq_deferred, txq); + TASK_INIT(&txq->intrtask, 0, xn_txq_tq_intr, txq); + + txq->tq = taskqueue_create_fast(txq->name, M_WAITOK, + taskqueue_thread_enqueue, &txq->tq); + + error = taskqueue_start_threads(&txq->tq, 1, PI_NET, + "%s txq %d", device_get_nameunit(dev), txq->id); + if (error != 0) { + device_printf(dev, "failed to start tx taskq %d\n", + txq->id); + goto fail_start_thread; + } + + error = xen_intr_alloc_and_bind_local_port(dev, + xenbus_get_otherend_id(dev), xn_intr, /* handler */ NULL, + &info->txq[q], + INTR_TYPE_NET | INTR_MPSAFE | INTR_ENTROPY, + &txq->xen_intr_handle); + + if (error != 0) { + device_printf(dev, "xen_intr_alloc_and_bind_local_port failed\n"); + goto fail_bind_port; + } + } + + return (0); + +fail_bind_port: + taskqueue_drain_all(txq->tq); +fail_start_thread: + gnttab_free_grant_references(txq->gref_head); + free(txq->ring.sring, M_DEVBUF); + gnttab_end_foreign_access_ref(txq->ring_ref); + buf_ring_free(txq->br, M_DEVBUF); + taskqueue_free(txq->tq); +fail_grant_ring: + gnttab_free_grant_references(txq->gref_head); + free(txq->ring.sring, M_DEVBUF); +fail: + for (; q >= 0; q--) { + disconnect_txq(&info->txq[q]); + destroy_txq(&info->txq[q]); + } + + free(info->txq, M_DEVBUF); + return (error); +} + +static int +setup_device(device_t dev, struct netfront_info *info, + unsigned long num_queues) +{ + int error; + int q; + + if (info->txq) + destroy_txqs(info); + + if (info->rxq) + destroy_rxqs(info); + + info->num_queues = 0; + + error = setup_rxqs(dev, info, num_queues); + if (error != 0) + goto out; + error = setup_txqs(dev, info, num_queues); + if (error != 0) + goto out; + + info->num_queues = num_queues; + + /* No split event channel at the moment. */ + for (q = 0; q < num_queues; q++) + info->rxq[q].xen_intr_handle = info->txq[q].xen_intr_handle; + + return (0); + +out: + KASSERT(error != 0, ("Error path taken without providing an error code")); return (error); } @@ -614,7 +1000,7 @@ netfront_backend_changed(device_t dev, XenbusState newstate) case XenbusStateInitWait: if (xenbus_get_state(dev) != XenbusStateInitialising) break; - if (network_connect(sc) != 0) + if (xn_connect(sc) != 0) break; xenbus_set_state(dev, XenbusStateConnected); break; @@ -629,42 +1015,6 @@ netfront_backend_changed(device_t dev, XenbusState newstate) } } -static void -xn_free_rx_ring(struct netfront_info *sc) -{ -#if 0 - int i; - - for (i = 0; i < NET_RX_RING_SIZE; i++) { - if (sc->xn_cdata.rx_mbufs[i] != NULL) { - m_freem(sc->rx_mbufs[i]); - sc->rx_mbufs[i] = NULL; - } - } - - sc->rx.rsp_cons = 0; - sc->xn_rx_if->req_prod = 0; - sc->xn_rx_if->event = sc->rx.rsp_cons ; -#endif -} - -static void -xn_free_tx_ring(struct netfront_info *sc) -{ -#if 0 - int i; - - for (i = 0; i < NET_TX_RING_SIZE; i++) { - if (sc->tx_mbufs[i] != NULL) { - m_freem(sc->tx_mbufs[i]); - sc->xn_cdata.xn_tx_chain[i] = NULL; - } - } - - return; -#endif -} - /** * \brief Verify that there is sufficient space in the Tx ring * buffer for a maximally sized request to be enqueued. @@ -673,20 +1023,21 @@ xn_free_tx_ring(struct netfront_info *sc) * fragment, plus up to 2 entries for "options" (e.g. TSO). */ static inline int -xn_tx_slot_available(struct netfront_info *np) +xn_tx_slot_available(struct netfront_txq *txq) { - return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2)); + + return (RING_FREE_REQUESTS(&txq->ring) > (MAX_TX_REQ_FRAGS + 2)); } static void -netif_release_tx_bufs(struct netfront_info *np) +xn_release_tx_bufs(struct netfront_txq *txq) { int i; for (i = 1; i <= NET_TX_RING_SIZE; i++) { struct mbuf *m; - m = np->tx_mbufs[i]; + m = txq->mbufs[i]; /* * We assume that no kernel addresses are @@ -696,13 +1047,13 @@ netif_release_tx_bufs(struct netfront_info *np) */ if (((uintptr_t)m) <= NET_TX_RING_SIZE) continue; - gnttab_end_foreign_access_ref(np->grant_tx_ref[i]); - gnttab_release_grant_reference(&np->gref_tx_head, - np->grant_tx_ref[i]); - np->grant_tx_ref[i] = GRANT_REF_INVALID; - add_id_to_freelist(np->tx_mbufs, i); - np->xn_cdata.xn_tx_chain_cnt--; - if (np->xn_cdata.xn_tx_chain_cnt < 0) { + gnttab_end_foreign_access_ref(txq->grant_ref[i]); + gnttab_release_grant_reference(&txq->gref_head, + txq->grant_ref[i]); + txq->grant_ref[i] = GRANT_REF_INVALID; + add_id_to_freelist(txq->mbufs, i); + txq->mbufs_cnt--; + if (txq->mbufs_cnt < 0) { panic("%s: tx_chain_cnt must be >= 0", __func__); } m_free(m); @@ -710,9 +1061,10 @@ netif_release_tx_bufs(struct netfront_info *np) } static void -network_alloc_rx_buffers(struct netfront_info *sc) +xn_alloc_rx_buffers(struct netfront_rxq *rxq) { - int otherend_id = xenbus_get_otherend_id(sc->xbdev); + struct netfront_info *np = rxq->info; + int otherend_id = xenbus_get_otherend_id(np->xbdev); unsigned short id; struct mbuf *m_new; int i, batch_target, notify; @@ -722,9 +1074,9 @@ network_alloc_rx_buffers(struct netfront_info *sc) vm_offset_t vaddr; u_long pfn; - req_prod = sc->rx.req_prod_pvt; + req_prod = rxq->ring.req_prod_pvt; - if (__predict_false(sc->carrier == 0)) + if (__predict_false(np->carrier == 0)) return; /* @@ -736,21 +1088,19 @@ network_alloc_rx_buffers(struct netfront_info *sc) * Here we attempt to maintain rx_target buffers in flight, counting * buffers that we have yet to process in the receive ring. */ - batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons); - for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) { + batch_target = rxq->target - (req_prod - rxq->ring.rsp_cons); + for (i = mbufq_len(&rxq->batch); i < batch_target; i++) { m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUMPAGESIZE); if (m_new == NULL) { if (i != 0) goto refill; - /* - * XXX set timer - */ + /* XXX set timer */ break; } m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE; /* queue the mbufs allocated */ - (void )mbufq_enqueue(&sc->xn_rx_batch, m_new); + mbufq_enqueue(&rxq->batch, m_new); } /* @@ -759,8 +1109,8 @@ network_alloc_rx_buffers(struct netfront_info *sc) * of submission worthwhile. Otherwise wait for more mbufs and * request entries to become available. */ - if (i < (sc->rx_target/2)) { - if (req_prod >sc->rx.sring->req_prod) + if (i < (rxq->target/2)) { + if (req_prod > rxq->ring.sring->req_prod) goto push; return; } @@ -771,44 +1121,44 @@ network_alloc_rx_buffers(struct netfront_info *sc) * low" as having less than a fourth of our target buffers free * at the time we refilled the queue. */ - if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) { - sc->rx_target *= 2; - if (sc->rx_target > sc->rx_max_target) - sc->rx_target = sc->rx_max_target; + if ((req_prod - rxq->ring.sring->rsp_prod) < (rxq->target / 4)) { + rxq->target *= 2; + if (rxq->target > np->rx_max_target) + rxq->target = np->rx_max_target; } refill: for (i = 0; ; i++) { - if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL) + if ((m_new = mbufq_dequeue(&rxq->batch)) == NULL) break; m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)( vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT); - id = xennet_rxidx(req_prod + i); + id = xn_rxidx(req_prod + i); - KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain")); - sc->rx_mbufs[id] = m_new; + KASSERT(rxq->mbufs[id] == NULL, ("non-NULL xn_rx_chain")); + rxq->mbufs[id] = m_new; - ref = gnttab_claim_grant_reference(&sc->gref_rx_head); + ref = gnttab_claim_grant_reference(&rxq->gref_head); KASSERT(ref != GNTTAB_LIST_END, ("reserved grant references exhuasted")); - sc->grant_rx_ref[id] = ref; + rxq->grant_ref[id] = ref; vaddr = mtod(m_new, vm_offset_t); pfn = vtophys(vaddr) >> PAGE_SHIFT; - req = RING_GET_REQUEST(&sc->rx, req_prod + i); + req = RING_GET_REQUEST(&rxq->ring, req_prod + i); gnttab_grant_foreign_access_ref(ref, otherend_id, pfn, 0); req->id = id; req->gref = ref; - sc->rx_pfn_array[i] = + rxq->pfn_array[i] = vtophys(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT; } KASSERT(i, ("no mbufs processed")); /* should have returned earlier */ - KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed")); + KASSERT(mbufq_len(&rxq->batch) == 0, ("not all mbufs processed")); /* * We may have allocated buffers which have entries outstanding * in the page * update queue -- make sure we flush those first! @@ -816,19 +1166,44 @@ network_alloc_rx_buffers(struct netfront_info *sc) wmb(); /* Above is a suitable barrier to ensure backend will see requests. */ - sc->rx.req_prod_pvt = req_prod + i; + rxq->ring.req_prod_pvt = req_prod + i; push: - RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify); + RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rxq->ring, notify); if (notify) - xen_intr_signal(sc->xen_intr_handle); + xen_intr_signal(rxq->xen_intr_handle); } static void -xn_rxeof(struct netfront_info *np) +xn_release_rx_bufs(struct netfront_rxq *rxq) +{ + int i, ref; + struct mbuf *m; + + for (i = 0; i < NET_RX_RING_SIZE; i++) { + m = rxq->mbufs[i]; + + if (m == NULL) + continue; + + ref = rxq->grant_ref[i]; + if (ref == GRANT_REF_INVALID) + continue; + + gnttab_end_foreign_access_ref(ref); + gnttab_release_grant_reference(&rxq->gref_head, ref); + rxq->mbufs[i] = NULL; + rxq->grant_ref[i] = GRANT_REF_INVALID; + m_freem(m); + } +} + +static void +xn_rxeof(struct netfront_rxq *rxq) { struct ifnet *ifp; + struct netfront_info *np = rxq->info; #if (defined(INET) || defined(INET6)) - struct lro_ctrl *lro = &np->xn_lro; + struct lro_ctrl *lro = &rxq->lro; struct lro_entry *queued; #endif struct netfront_rx_info rinfo; @@ -836,35 +1211,35 @@ xn_rxeof(struct netfront_info *np) struct netif_extra_info *extras = rinfo.extras; RING_IDX i, rp; struct mbuf *m; - struct mbufq rxq, errq; + struct mbufq mbufq_rxq, mbufq_errq; int err, work_to_do; do { - XN_RX_LOCK_ASSERT(np); + XN_RX_LOCK_ASSERT(rxq); if (!netfront_carrier_ok(np)) return; /* XXX: there should be some sane limit. */ - mbufq_init(&errq, INT_MAX); - mbufq_init(&rxq, INT_MAX); + mbufq_init(&mbufq_errq, INT_MAX); + mbufq_init(&mbufq_rxq, INT_MAX); ifp = np->xn_ifp; - rp = np->rx.sring->rsp_prod; + rp = rxq->ring.sring->rsp_prod; rmb(); /* Ensure we see queued responses up to 'rp'. */ - i = np->rx.rsp_cons; + i = rxq->ring.rsp_cons; while ((i != rp)) { - memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); + memcpy(rx, RING_GET_RESPONSE(&rxq->ring, i), sizeof(*rx)); memset(extras, 0, sizeof(rinfo.extras)); m = NULL; - err = xennet_get_responses(np, &rinfo, rp, &i, &m); + err = xn_get_responses(rxq, &rinfo, rp, &i, &m); if (__predict_false(err)) { if (m) - (void )mbufq_enqueue(&errq, m); - np->stats.rx_errors++; + (void )mbufq_enqueue(&mbufq_errq, m); + rxq->stats.rx_errors++; continue; } @@ -882,26 +1257,24 @@ xn_rxeof(struct netfront_info *np) m->m_pkthdr.csum_data = 0xffff; } - np->stats.rx_packets++; - np->stats.rx_bytes += m->m_pkthdr.len; + rxq->stats.rx_packets++; + rxq->stats.rx_bytes += m->m_pkthdr.len; - (void )mbufq_enqueue(&rxq, m); - np->rx.rsp_cons = i; + (void )mbufq_enqueue(&mbufq_rxq, m); + rxq->ring.rsp_cons = i; } - mbufq_drain(&errq); + mbufq_drain(&mbufq_errq); /* * Process all the mbufs after the remapping is complete. * Break the mbuf chain first though. */ - while ((m = mbufq_dequeue(&rxq)) != NULL) { + while ((m = mbufq_dequeue(&mbufq_rxq)) != NULL) { if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); - /* - * Do we really need to drop the rx lock? - */ - XN_RX_UNLOCK(np); + /* XXX: Do we really need to drop the rx lock? */ + XN_RX_UNLOCK(rxq); #if (defined(INET) || defined(INET6)) /* Use LRO if possible */ if ((ifp->if_capenable & IFCAP_LRO) == 0 || @@ -915,10 +1288,11 @@ xn_rxeof(struct netfront_info *np) #else (*ifp->if_input)(ifp, m); #endif - XN_RX_LOCK(np); + + XN_RX_LOCK(rxq); } - np->rx.rsp_cons = i; + rxq->ring.rsp_cons = i; #if (defined(INET) || defined(INET6)) /* @@ -931,30 +1305,23 @@ xn_rxeof(struct netfront_info *np) } #endif -#if 0 - /* If we get a callback with very few responses, reduce fill target. */ - /* NB. Note exponential increase, linear decrease. */ - if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > - ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target)) - np->rx_target = np->rx_min_target; -#endif + xn_alloc_rx_buffers(rxq); - network_alloc_rx_buffers(np); - - RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do); + RING_FINAL_CHECK_FOR_RESPONSES(&rxq->ring, work_to_do); } while (work_to_do); } static void -xn_txeof(struct netfront_info *np) +xn_txeof(struct netfront_txq *txq) { RING_IDX i, prod; unsigned short id; struct ifnet *ifp; netif_tx_response_t *txr; struct mbuf *m; + struct netfront_info *np = txq->info; - XN_TX_LOCK_ASSERT(np); + XN_TX_LOCK_ASSERT(txq); if (!netfront_carrier_ok(np)) return; @@ -962,11 +1329,11 @@ xn_txeof(struct netfront_info *np) ifp = np->xn_ifp; do { - prod = np->tx.sring->rsp_prod; + prod = txq->ring.sring->rsp_prod; rmb(); /* Ensure we see responses up to 'rp'. */ - for (i = np->tx.rsp_cons; i != prod; i++) { - txr = RING_GET_RESPONSE(&np->tx, i); + for (i = txq->ring.rsp_cons; i != prod; i++) { + txr = RING_GET_RESPONSE(&txq->ring, i); if (txr->status == NETIF_RSP_NULL) continue; @@ -975,8 +1342,8 @@ xn_txeof(struct netfront_info *np) __func__, txr->status); } id = txr->id; - m = np->tx_mbufs[id]; - KASSERT(m != NULL, ("mbuf not found in xn_tx_chain")); + m = txq->mbufs[id]; + KASSERT(m != NULL, ("mbuf not found in chain")); KASSERT((uintptr_t)m > NET_TX_RING_SIZE, ("mbuf already on the free list, but we're " "trying to free it again!")); @@ -989,24 +1356,23 @@ xn_txeof(struct netfront_info *np) if (!m->m_next) if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if (__predict_false(gnttab_query_foreign_access( - np->grant_tx_ref[id]) != 0)) { + txq->grant_ref[id]) != 0)) { panic("%s: grant id %u still in use by the " "backend", __func__, id); } - gnttab_end_foreign_access_ref( - np->grant_tx_ref[id]); + gnttab_end_foreign_access_ref(txq->grant_ref[id]); gnttab_release_grant_reference( - &np->gref_tx_head, np->grant_tx_ref[id]); - np->grant_tx_ref[id] = GRANT_REF_INVALID; + &txq->gref_head, txq->grant_ref[id]); + txq->grant_ref[id] = GRANT_REF_INVALID; - np->tx_mbufs[id] = NULL; - add_id_to_freelist(np->tx_mbufs, id); - np->xn_cdata.xn_tx_chain_cnt--; + txq->mbufs[id] = NULL; + add_id_to_freelist(txq->mbufs, id); + txq->mbufs_cnt--; m_free(m); - /* Only mark the queue active if we've freed up at least one slot to try */ + /* Only mark the txq active if we've freed up at least one slot to try */ ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; } - np->tx.rsp_cons = prod; + txq->ring.rsp_cons = prod; /* * Set a new event, then check for race with update of @@ -1017,65 +1383,66 @@ xn_txeof(struct netfront_info *np) * cases notification from Xen is likely to be the only kick * that we'll get. */ - np->tx.sring->rsp_event = - prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; + txq->ring.sring->rsp_event = + prod + ((txq->ring.sring->req_prod - prod) >> 1) + 1; mb(); - } while (prod != np->tx.sring->rsp_prod); + } while (prod != txq->ring.sring->rsp_prod); - if (np->tx_full && - ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) { - np->tx_full = 0; -#if 0 - if (np->user_state == UST_OPEN) - netif_wake_queue(dev); -#endif + if (txq->full && + ((txq->ring.sring->req_prod - prod) < NET_TX_RING_SIZE)) { + txq->full = false; + taskqueue_enqueue(txq->tq, &txq->intrtask); } } + static void -xn_intr(void *xsc) +xn_rxq_intr(void *xrxq) { - struct netfront_info *np = xsc; - struct ifnet *ifp = np->xn_ifp; + struct netfront_rxq *rxq = xrxq; -#if 0 - if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod && - likely(netfront_carrier_ok(np)) && - ifp->if_drv_flags & IFF_DRV_RUNNING)) - return; -#endif - if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) { - XN_TX_LOCK(np); - xn_txeof(np); - XN_TX_UNLOCK(np); - } - - XN_RX_LOCK(np); - xn_rxeof(np); - XN_RX_UNLOCK(np); - - if (ifp->if_drv_flags & IFF_DRV_RUNNING && - !IFQ_DRV_IS_EMPTY(&ifp->if_snd)) - xn_start(ifp); + taskqueue_enqueue_fast(rxq->tq, &rxq->intrtask); } static void -xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m, - grant_ref_t ref) +xn_txq_intr(void *xtxq) { - int new = xennet_rxidx(np->rx.req_prod_pvt); + struct netfront_txq *txq = xtxq; - KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL")); - np->rx_mbufs[new] = m; - np->grant_rx_ref[new] = ref; - RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; - RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; - np->rx.req_prod_pvt++; + taskqueue_enqueue_fast(txq->tq, &txq->intrtask); } static int -xennet_get_extras(struct netfront_info *np, +xn_intr(void *xsc) +{ + struct netfront_txq *txq = xsc; + struct netfront_info *np = txq->info; + struct netfront_rxq *rxq = &np->rxq[txq->id]; + + /* kick both tx and rx */ + xn_rxq_intr(rxq); + xn_txq_intr(txq); + + return (FILTER_HANDLED); +} + +static void +xn_move_rx_slot(struct netfront_rxq *rxq, struct mbuf *m, + grant_ref_t ref) +{ + int new = xn_rxidx(rxq->ring.req_prod_pvt); + + KASSERT(rxq->mbufs[new] == NULL, ("mbufs != NULL")); + rxq->mbufs[new] = m; + rxq->grant_ref[new] = ref; + RING_GET_REQUEST(&rxq->ring, rxq->ring.req_prod_pvt)->id = new; + RING_GET_REQUEST(&rxq->ring, rxq->ring.req_prod_pvt)->gref = ref; + rxq->ring.req_prod_pvt++; +} + +static int +xn_get_extras(struct netfront_rxq *rxq, struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons) { struct netif_extra_info *extra; @@ -1087,55 +1454,46 @@ xennet_get_extras(struct netfront_info *np, grant_ref_t ref; if (__predict_false(*cons + 1 == rp)) { -#if 0 - if (net_ratelimit()) - WPRINTK("Missing extra info\n"); -#endif err = EINVAL; break; } extra = (struct netif_extra_info *) - RING_GET_RESPONSE(&np->rx, ++(*cons)); + RING_GET_RESPONSE(&rxq->ring, ++(*cons)); if (__predict_false(!extra->type || extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { -#if 0 - if (net_ratelimit()) - WPRINTK("Invalid extra type: %d\n", - extra->type); -#endif err = EINVAL; } else { memcpy(&extras[extra->type - 1], extra, sizeof(*extra)); } - m = xennet_get_rx_mbuf(np, *cons); - ref = xennet_get_rx_ref(np, *cons); - xennet_move_rx_slot(np, m, ref); + m = xn_get_rx_mbuf(rxq, *cons); + ref = xn_get_rx_ref(rxq, *cons); + xn_move_rx_slot(rxq, m, ref); } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); return err; } static int -xennet_get_responses(struct netfront_info *np, - struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons, - struct mbuf **list) +xn_get_responses(struct netfront_rxq *rxq, + struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons, + struct mbuf **list) { struct netif_rx_response *rx = &rinfo->rx; struct netif_extra_info *extras = rinfo->extras; struct mbuf *m, *m0, *m_prev; - grant_ref_t ref = xennet_get_rx_ref(np, *cons); + grant_ref_t ref = xn_get_rx_ref(rxq, *cons); RING_IDX ref_cons = *cons; int frags = 1; int err = 0; u_long ret; - m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons); + m0 = m = m_prev = xn_get_rx_mbuf(rxq, *cons); if (rx->flags & NETRXF_extra_info) { - err = xennet_get_extras(np, extras, rp, cons); + err = xn_get_extras(rxq, extras, rp, cons); } if (m0 != NULL) { @@ -1151,12 +1509,7 @@ xennet_get_responses(struct netfront_info *np, if (__predict_false(rx->status < 0 || rx->offset + rx->status > PAGE_SIZE)) { -#if 0 - if (net_ratelimit()) - WPRINTK("rx->offset: %x, size: %u\n", - rx->offset, rx->status); -#endif - xennet_move_rx_slot(np, m, ref); + xn_move_rx_slot(rxq, m, ref); if (m0 == m) m0 = NULL; m = NULL; @@ -1170,12 +1523,7 @@ xennet_get_responses(struct netfront_info *np, * situation to the system controller to reboot the backed. */ if (ref == GRANT_REF_INVALID) { - -#if 0 - if (net_ratelimit()) - WPRINTK("Bad rx response id %d.\n", rx->id); -#endif - printf("%s: Bad rx response id %d.\n", __func__,rx->id); + printf("%s: Bad rx response id %d.\n", __func__, rx->id); err = EINVAL; goto next; } @@ -1183,7 +1531,7 @@ xennet_get_responses(struct netfront_info *np, ret = gnttab_end_foreign_access_ref(ref); KASSERT(ret, ("Unable to end access to grant references")); - gnttab_release_grant_reference(&np->gref_rx_head, ref); + gnttab_release_grant_reference(&rxq->gref_head, ref); next: if (m == NULL) @@ -1211,8 +1559,8 @@ xennet_get_responses(struct netfront_info *np, */ m_prev = m; - rx = RING_GET_RESPONSE(&np->rx, *cons + frags); - m = xennet_get_rx_mbuf(np, *cons + frags); + rx = RING_GET_RESPONSE(&rxq->ring, *cons + frags); + m = xn_get_rx_mbuf(rxq, *cons + frags); /* * m_prev == NULL can happen if rx->status < 0 or if @@ -1228,7 +1576,7 @@ xennet_get_responses(struct netfront_info *np, if (m0 == NULL) m0 = m; m->m_next = NULL; - ref = xennet_get_rx_ref(np, *cons + frags); + ref = xn_get_rx_ref(rxq, *cons + frags); ref_cons = *cons + frags; frags++; } @@ -1238,26 +1586,6 @@ xennet_get_responses(struct netfront_info *np, return (err); } -static void -xn_tick_locked(struct netfront_info *sc) -{ - XN_RX_LOCK_ASSERT(sc); - callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc); - - /* XXX placeholder for printing debug information */ -} - -static void -xn_tick(void *xsc) -{ - struct netfront_info *sc; - - sc = xsc; - XN_RX_LOCK(sc); - xn_tick_locked(sc); - XN_RX_UNLOCK(sc); -} - /** * \brief Count the number of fragments in an mbuf chain. * @@ -1279,15 +1607,14 @@ xn_count_frags(struct mbuf *m) * it onto the transmit ring. */ static int -xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head) +xn_assemble_tx_request(struct netfront_txq *txq, struct mbuf *m_head) { - struct ifnet *ifp; struct mbuf *m; + struct netfront_info *np = txq->info; + struct ifnet *ifp = np->xn_ifp; u_int nfrags; int otherend_id; - ifp = sc->xn_ifp; - /** * Defragment the mbuf if necessary. */ @@ -1302,7 +1629,7 @@ xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head) * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of * the Linux network stack. */ - if (nfrags > sc->maxfrags) { + if (nfrags > np->maxfrags) { m = m_defrag(m_head, M_NOWAIT); if (!m) { /* @@ -1344,11 +1671,11 @@ xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head) * have enough slots in the ring to handle a packet of maximum * size, and that our packet is less than the maximum size. Keep * it in here as an assert for now just to make certain that - * xn_tx_chain_cnt is accurate. + * chain_cnt is accurate. */ - KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE, - ("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE " - "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt, + KASSERT((txq->mbufs_cnt + nfrags) <= NET_TX_RING_SIZE, + ("%s: chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE " + "(%d)!", __func__, (int) txq->mbufs_cnt, (int) nfrags, (int) NET_TX_RING_SIZE)); /* @@ -1357,30 +1684,30 @@ xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head) * of fragments or hit the end of the mbuf chain. */ m = m_head; - otherend_id = xenbus_get_otherend_id(sc->xbdev); + otherend_id = xenbus_get_otherend_id(np->xbdev); for (m = m_head; m; m = m->m_next) { netif_tx_request_t *tx; uintptr_t id; grant_ref_t ref; u_long mfn; /* XXX Wrong type? */ - tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt); - id = get_id_from_freelist(sc->tx_mbufs); + tx = RING_GET_REQUEST(&txq->ring, txq->ring.req_prod_pvt); + id = get_id_from_freelist(txq->mbufs); if (id == 0) panic("%s: was allocated the freelist head!\n", __func__); - sc->xn_cdata.xn_tx_chain_cnt++; - if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE) + txq->mbufs_cnt++; + if (txq->mbufs_cnt > NET_TX_RING_SIZE) panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n", __func__); - sc->tx_mbufs[id] = m; + txq->mbufs[id] = m; tx->id = id; - ref = gnttab_claim_grant_reference(&sc->gref_tx_head); + ref = gnttab_claim_grant_reference(&txq->gref_head); KASSERT((short)ref >= 0, ("Negative ref")); mfn = virt_to_mfn(mtod(m, vm_offset_t)); gnttab_grant_foreign_access_ref(ref, otherend_id, mfn, GNTMAP_readonly); - tx->gref = sc->grant_tx_ref[id] = ref; + tx->gref = txq->grant_ref[id] = ref; tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1); tx->flags = 0; if (m == m_head) { @@ -1414,8 +1741,8 @@ xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head) if (m->m_pkthdr.csum_flags & CSUM_TSO) { struct netif_extra_info *gso = (struct netif_extra_info *) - RING_GET_REQUEST(&sc->tx, - ++sc->tx.req_prod_pvt); + RING_GET_REQUEST(&txq->ring, + ++txq->ring.req_prod_pvt); tx->flags |= NETTXF_extra_info; @@ -1434,87 +1761,44 @@ xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head) if (m->m_next) tx->flags |= NETTXF_more_data; - sc->tx.req_prod_pvt++; + txq->ring.req_prod_pvt++; } BPF_MTAP(ifp, m_head); - sc->stats.tx_bytes += m_head->m_pkthdr.len; - sc->stats.tx_packets++; + xn_txeof(txq); + + txq->stats.tx_bytes += m_head->m_pkthdr.len; + txq->stats.tx_packets++; return (0); } -static void -xn_start_locked(struct ifnet *ifp) -{ - struct netfront_info *sc; - struct mbuf *m_head; - int notify; - - sc = ifp->if_softc; - - if (!netfront_carrier_ok(sc)) - return; - - /* - * While we have enough transmit slots available for at least one - * maximum-sized packet, pull mbufs off the queue and put them on - * the transmit ring. - */ - while (xn_tx_slot_available(sc)) { - IF_DEQUEUE(&ifp->if_snd, m_head); - if (m_head == NULL) - break; - - if (xn_assemble_tx_request(sc, m_head) != 0) - break; - } - - RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify); - if (notify) - xen_intr_signal(sc->xen_intr_handle); - - if (RING_FULL(&sc->tx)) { - sc->tx_full = 1; -#if 0 - netif_stop_queue(dev); -#endif - } -} - -static void -xn_start(struct ifnet *ifp) -{ - struct netfront_info *sc; - sc = ifp->if_softc; - XN_TX_LOCK(sc); - xn_start_locked(ifp); - XN_TX_UNLOCK(sc); -} - /* equivalent of network_open() in Linux */ static void -xn_ifinit_locked(struct netfront_info *sc) +xn_ifinit_locked(struct netfront_info *np) { struct ifnet *ifp; + int i; + struct netfront_rxq *rxq; - XN_LOCK_ASSERT(sc); + XN_LOCK_ASSERT(np); - ifp = sc->xn_ifp; + ifp = np->xn_ifp; if (ifp->if_drv_flags & IFF_DRV_RUNNING) return; - xn_stop(sc); + xn_stop(np); - network_alloc_rx_buffers(sc); - sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1; + for (i = 0; i < np->num_queues; i++) { + rxq = &np->rxq[i]; + xn_alloc_rx_buffers(rxq); + rxq->ring.sring->rsp_event = rxq->ring.rsp_cons + 1; + } ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; if_link_state_change(ifp, LINK_STATE_UP); - - callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc); } static void @@ -1556,17 +1840,9 @@ xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) #endif break; case SIOCSIFMTU: - /* XXX can we alter the MTU on a VN ?*/ -#ifdef notyet - if (ifr->ifr_mtu > XN_JUMBO_MTU) - error = EINVAL; - else -#endif - { - ifp->if_mtu = ifr->ifr_mtu; - ifp->if_drv_flags &= ~IFF_DRV_RUNNING; - xn_ifinit(sc); - } + ifp->if_mtu = ifr->ifr_mtu; + ifp->if_drv_flags &= ~IFF_DRV_RUNNING; + xn_ifinit(sc); break; case SIOCSIFFLAGS: XN_LOCK(sc); @@ -1579,21 +1855,7 @@ xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) * waiting for it to start up, which may take a * second or two. */ -#ifdef notyet - /* No promiscuous mode with Xen */ - if (ifp->if_drv_flags & IFF_DRV_RUNNING && - ifp->if_flags & IFF_PROMISC && - !(sc->xn_if_flags & IFF_PROMISC)) { - XN_SETBIT(sc, XN_RX_MODE, - XN_RXMODE_RX_PROMISC); - } else if (ifp->if_drv_flags & IFF_DRV_RUNNING && - !(ifp->if_flags & IFF_PROMISC) && - sc->xn_if_flags & IFF_PROMISC) { - XN_CLRBIT(sc, XN_RX_MODE, - XN_RXMODE_RX_PROMISC); - } else -#endif - xn_ifinit_locked(sc); + xn_ifinit_locked(sc); } else { if (ifp->if_drv_flags & IFF_DRV_RUNNING) { xn_stop(sc); @@ -1640,14 +1902,6 @@ xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) break; case SIOCADDMULTI: case SIOCDELMULTI: -#ifdef notyet - if (ifp->if_drv_flags & IFF_DRV_RUNNING) { - XN_LOCK(sc); - xn_setmulti(sc); - XN_UNLOCK(sc); - error = 0; - } -#endif break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: @@ -1669,59 +1923,32 @@ xn_stop(struct netfront_info *sc) ifp = sc->xn_ifp; - callout_stop(&sc->xn_stat_ch); - - xn_free_rx_ring(sc); - xn_free_tx_ring(sc); - ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); if_link_state_change(ifp, LINK_STATE_DOWN); } -/* START of Xenolinux helper functions adapted to FreeBSD */ -int -network_connect(struct netfront_info *np) +static void +xn_rebuild_rx_bufs(struct netfront_rxq *rxq) { - int i, requeue_idx, error; + int requeue_idx, i; grant_ref_t ref; netif_rx_request_t *req; - u_int feature_rx_copy; - error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), - "feature-rx-copy", NULL, "%u", &feature_rx_copy); - if (error) - feature_rx_copy = 0; - - /* We only support rx copy. */ - if (!feature_rx_copy) - return (EPROTONOSUPPORT); - - /* Recovery procedure: */ - error = talk_to_backend(np->xbdev, np); - if (error) - return (error); - - /* Step 1: Reinitialise variables. */ - xn_query_features(np); - xn_configure_features(np); - netif_release_tx_bufs(np); - - /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { struct mbuf *m; u_long pfn; - if (np->rx_mbufs[i] == NULL) + if (rxq->mbufs[i] == NULL) continue; - m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i); - ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); + m = rxq->mbufs[requeue_idx] = xn_get_rx_mbuf(rxq, i); + ref = rxq->grant_ref[requeue_idx] = xn_get_rx_ref(rxq, i); - req = RING_GET_REQUEST(&np->rx, requeue_idx); + req = RING_GET_REQUEST(&rxq->ring, requeue_idx); pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT; gnttab_grant_foreign_access_ref(ref, - xenbus_get_otherend_id(np->xbdev), + xenbus_get_otherend_id(rxq->info->xbdev), pfn, 0); req->gref = ref; @@ -1730,19 +1957,62 @@ network_connect(struct netfront_info *np) requeue_idx++; } - np->rx.req_prod_pvt = requeue_idx; + rxq->ring.req_prod_pvt = requeue_idx; +} - /* Step 3: All public and private state should now be sane. Get +/* START of Xenolinux helper functions adapted to FreeBSD */ +int +xn_connect(struct netfront_info *np) +{ + int i, error; + u_int feature_rx_copy; + struct netfront_rxq *rxq; + struct netfront_txq *txq; + + error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), + "feature-rx-copy", NULL, "%u", &feature_rx_copy); + if (error != 0) + feature_rx_copy = 0; + + /* We only support rx copy. */ + if (!feature_rx_copy) + return (EPROTONOSUPPORT); + + /* Recovery procedure: */ + error = talk_to_backend(np->xbdev, np); + if (error != 0) + return (error); + + /* Step 1: Reinitialise variables. */ + xn_query_features(np); + xn_configure_features(np); + + /* Step 2: Release TX buffer */ + for (i = 0; i < np->num_queues; i++) { + txq = &np->txq[i]; + xn_release_tx_bufs(txq); + } + + /* Step 3: Rebuild the RX buffer freelist and the RX ring itself. */ + for (i = 0; i < np->num_queues; i++) { + rxq = &np->rxq[i]; + xn_rebuild_rx_bufs(rxq); + } + + /* Step 4: All public and private state should now be sane. Get * ready to start sending and receiving packets and give the driver * domain a kick because we've probably just requeued some * packets. */ netfront_carrier_on(np); - xen_intr_signal(np->xen_intr_handle); - XN_TX_LOCK(np); - xn_txeof(np); - XN_TX_UNLOCK(np); - network_alloc_rx_buffers(np); + for (i = 0; i < np->num_queues; i++) { + txq = &np->txq[i]; + xen_intr_signal(txq->xen_intr_handle); + XN_TX_LOCK(txq); + xn_txeof(txq); + XN_TX_UNLOCK(txq); + xn_alloc_rx_buffers(rxq); + } return (0); } @@ -1781,6 +2051,9 @@ static int xn_configure_features(struct netfront_info *np) { int err, cap_enabled; +#if (defined(INET) || defined(INET6)) + int i; +#endif err = 0; @@ -1798,21 +2071,25 @@ xn_configure_features(struct netfront_info *np) cap_enabled = UINT_MAX; #if (defined(INET) || defined(INET6)) - if ((np->xn_ifp->if_capenable & IFCAP_LRO) == (cap_enabled & IFCAP_LRO)) - tcp_lro_free(&np->xn_lro); + for (i = 0; i < np->num_queues; i++) + if ((np->xn_ifp->if_capenable & IFCAP_LRO) == + (cap_enabled & IFCAP_LRO)) + tcp_lro_free(&np->rxq[i].lro); #endif np->xn_ifp->if_capenable = np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4) & cap_enabled; np->xn_ifp->if_hwassist &= ~CSUM_TSO; #if (defined(INET) || defined(INET6)) - if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) == - (cap_enabled & IFCAP_LRO)) { - err = tcp_lro_init(&np->xn_lro); - if (err) { - device_printf(np->xbdev, "LRO initialization failed\n"); - } else { - np->xn_lro.ifp = np->xn_ifp; - np->xn_ifp->if_capenable |= IFCAP_LRO; + for (i = 0; i < np->num_queues; i++) { + if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) == + (cap_enabled & IFCAP_LRO)) { + err = tcp_lro_init(&np->rxq[i].lro); + if (err != 0) { + device_printf(np->xbdev, "LRO initialization failed\n"); + } else { + np->rxq[i].lro.ifp = np->xn_ifp; + np->xn_ifp->if_capenable |= IFCAP_LRO; + } } } if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) == @@ -1824,6 +2101,111 @@ xn_configure_features(struct netfront_info *np) return (err); } +static int +xn_txq_mq_start_locked(struct netfront_txq *txq, struct mbuf *m) +{ + struct netfront_info *np; + struct ifnet *ifp; + struct buf_ring *br; + int error, notify; + + np = txq->info; + br = txq->br; + ifp = np->xn_ifp; + error = 0; + + XN_TX_LOCK_ASSERT(txq); + + if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || + !netfront_carrier_ok(np)) { + if (m != NULL) + error = drbr_enqueue(ifp, br, m); + return (error); + } + + if (m != NULL) { + error = drbr_enqueue(ifp, br, m); + if (error != 0) + return (error); + } + + while ((m = drbr_peek(ifp, br)) != NULL) { + if (!xn_tx_slot_available(txq)) { + drbr_putback(ifp, br, m); + break; + } + + error = xn_assemble_tx_request(txq, m); + /* xn_assemble_tx_request always consumes the mbuf*/ + if (error != 0) { + drbr_advance(ifp, br); + break; + } + + RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&txq->ring, notify); + if (notify) + xen_intr_signal(txq->xen_intr_handle); + + drbr_advance(ifp, br); + } + + if (RING_FULL(&txq->ring)) + txq->full = true; + + return (0); +} + +static int +xn_txq_mq_start(struct ifnet *ifp, struct mbuf *m) +{ + struct netfront_info *np; + struct netfront_txq *txq; + int i, npairs, error; + + np = ifp->if_softc; + npairs = np->num_queues; + + /* check if flowid is set */ + if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) + i = m->m_pkthdr.flowid % npairs; + else + i = curcpu % npairs; + + txq = &np->txq[i]; + + if (XN_TX_TRYLOCK(txq) != 0) { + error = xn_txq_mq_start_locked(txq, m); + XN_TX_UNLOCK(txq); + } else { + error = drbr_enqueue(ifp, txq->br, m); + taskqueue_enqueue(txq->tq, &txq->defrtask); + } + + return (error); +} + +static void +xn_qflush(struct ifnet *ifp) +{ + struct netfront_info *np; + struct netfront_txq *txq; + struct mbuf *m; + int i; + + np = ifp->if_softc; + + for (i = 0; i < np->num_queues; i++) { + txq = &np->txq[i]; + + XN_TX_LOCK(txq); + while ((m = buf_ring_dequeue_sc(txq->br)) != NULL) + m_freem(m); + XN_TX_UNLOCK(txq); + } + + if_qflush(ifp); +} + /** * Create a network device. * @param dev Newbus device representing this virtual NIC. @@ -1831,7 +2213,6 @@ xn_configure_features(struct netfront_info *np) int create_netdev(device_t dev) { - int i; struct netfront_info *np; int err; struct ifnet *ifp; @@ -1840,55 +2221,18 @@ create_netdev(device_t dev) np->xbdev = dev; - mtx_init(&np->tx_lock, "xntx", "netfront transmit lock", MTX_DEF); - mtx_init(&np->rx_lock, "xnrx", "netfront receive lock", MTX_DEF); mtx_init(&np->sc_lock, "xnsc", "netfront softc lock", MTX_DEF); ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts); ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL); ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL); - np->rx_target = RX_MIN_TARGET; np->rx_min_target = RX_MIN_TARGET; np->rx_max_target = RX_MAX_TARGET; - /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */ - for (i = 0; i <= NET_TX_RING_SIZE; i++) { - np->tx_mbufs[i] = (void *) ((u_long) i+1); - np->grant_tx_ref[i] = GRANT_REF_INVALID; - } - np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0; - - for (i = 0; i <= NET_RX_RING_SIZE; i++) { - - np->rx_mbufs[i] = NULL; - np->grant_rx_ref[i] = GRANT_REF_INVALID; - } - - mbufq_init(&np->xn_rx_batch, INT_MAX); - - /* A grant for every tx ring slot */ - if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, - &np->gref_tx_head) != 0) { - IPRINTK("#### netfront can't alloc tx grant refs\n"); - err = ENOMEM; - goto error; - } - /* A grant for every rx ring slot */ - if (gnttab_alloc_grant_references(RX_MAX_TARGET, - &np->gref_rx_head) != 0) { - WPRINTK("#### netfront can't alloc rx grant refs\n"); - gnttab_free_grant_references(np->gref_tx_head); - err = ENOMEM; - goto error; - } - err = xen_net_read_mac(dev, np->mac); - if (err) { - gnttab_free_grant_references(np->gref_rx_head); - gnttab_free_grant_references(np->gref_tx_head); + if (err != 0) goto error; - } /* Set up ifnet structure */ ifp = np->xn_ifp = if_alloc(IFT_ETHER); @@ -1896,12 +2240,11 @@ create_netdev(device_t dev) if_initname(ifp, "xn", device_get_unit(dev)); ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = xn_ioctl; - ifp->if_start = xn_start; -#ifdef notyet - ifp->if_watchdog = xn_watchdog; -#endif + + ifp->if_transmit = xn_txq_mq_start; + ifp->if_qflush = xn_qflush; + ifp->if_init = xn_ifinit; - ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1; ifp->if_hwassist = XN_CSUM_FEATURES; ifp->if_capabilities = IFCAP_HWCSUM; @@ -1910,7 +2253,6 @@ create_netdev(device_t dev) ifp->if_hw_tsomaxsegsize = PAGE_SIZE; ether_ifattach(ifp, np->mac); - callout_init(&np->xn_stat_ch, 1); netfront_carrier_off(np); return (0); @@ -1920,27 +2262,6 @@ create_netdev(device_t dev) return (err); } -/** - * Handle the change of state of the backend to Closing. We must delete our - * device-layer structures now, to ensure that writes are flushed through to - * the backend. Once is this done, we can switch to Closed in - * acknowledgement. - */ -#if 0 -static void -netfront_closing(device_t dev) -{ -#if 0 - struct netfront_info *info = dev->dev_driver_data; - - DPRINTK("netfront_closing: %s removed\n", dev->nodename); - - close_netdev(info); -#endif - xenbus_switch_state(dev, XenbusStateClosed); -} -#endif - static int netfront_detach(device_t dev) { @@ -1954,58 +2275,55 @@ netfront_detach(device_t dev) } static void -netif_free(struct netfront_info *info) +netif_free(struct netfront_info *np) { - XN_LOCK(info); - xn_stop(info); - XN_UNLOCK(info); - callout_drain(&info->xn_stat_ch); - netif_disconnect_backend(info); - if (info->xn_ifp != NULL) { - ether_ifdetach(info->xn_ifp); - if_free(info->xn_ifp); - info->xn_ifp = NULL; + + XN_LOCK(np); + xn_stop(np); + XN_UNLOCK(np); + netif_disconnect_backend(np); + free(np->rxq, M_DEVBUF); + free(np->txq, M_DEVBUF); + if (np->xn_ifp != NULL) { + ether_ifdetach(np->xn_ifp); + if_free(np->xn_ifp); + np->xn_ifp = NULL; } - ifmedia_removeall(&info->sc_media); + ifmedia_removeall(&np->sc_media); } static void -netif_disconnect_backend(struct netfront_info *info) +netif_disconnect_backend(struct netfront_info *np) { - XN_RX_LOCK(info); - XN_TX_LOCK(info); - netfront_carrier_off(info); - XN_TX_UNLOCK(info); - XN_RX_UNLOCK(info); + u_int i; - free_ring(&info->tx_ring_ref, &info->tx.sring); - free_ring(&info->rx_ring_ref, &info->rx.sring); - - xen_intr_unbind(&info->xen_intr_handle); -} - -static void -free_ring(int *ref, void *ring_ptr_ref) -{ - void **ring_ptr_ptr = ring_ptr_ref; - - if (*ref != GRANT_REF_INVALID) { - /* This API frees the associated storage. */ - gnttab_end_foreign_access(*ref, *ring_ptr_ptr); - *ref = GRANT_REF_INVALID; + for (i = 0; i < np->num_queues; i++) { + XN_RX_LOCK(&np->rxq[i]); + XN_TX_LOCK(&np->txq[i]); + } + netfront_carrier_off(np); + for (i = 0; i < np->num_queues; i++) { + XN_RX_UNLOCK(&np->rxq[i]); + XN_TX_UNLOCK(&np->txq[i]); + } + + for (i = 0; i < np->num_queues; i++) { + disconnect_rxq(&np->rxq[i]); + disconnect_txq(&np->txq[i]); } - *ring_ptr_ptr = NULL; } static int xn_ifmedia_upd(struct ifnet *ifp) { + return (0); } static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) { + ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE; ifmr->ifm_active = IFM_ETHER|IFM_MANUAL; }