X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev.c;h=284be38dab7b69eff9f78433578153d7fe863ec8;hb=8d25251929c8f325bed0fff24192d5a87034b32e;hp=bcdc5ee70c929a65177f2d6e24ecee3a2f316bd5;hpb=fe29af4c888d48cc1f16b1a247c2ffb6f0864522;p=sliver-openvswitch.git diff --git a/lib/netdev.c b/lib/netdev.c index bcdc5ee70..284be38da 100644 --- a/lib/netdev.c +++ b/lib/netdev.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc. + * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -31,6 +31,7 @@ #include "fatal-signal.h" #include "hash.h" #include "list.h" +#include "netdev-dpdk.h" #include "netdev-provider.h" #include "netdev-vport.h" #include "ofpbuf.h" @@ -67,15 +68,14 @@ static struct shash netdev_shash OVS_GUARDED_BY(netdev_mutex) /* Protects 'netdev_classes' against insertions or deletions. * - * This is not an rwlock for performance reasons but to allow recursive - * acquisition when calling into providers. For example, netdev_run() calls - * into provider 'run' functions, which might reasonably want to call one of - * the netdev functions that takes netdev_class_rwlock read-only. */ -static struct ovs_rwlock netdev_class_rwlock OVS_ACQ_BEFORE(netdev_mutex) - = OVS_RWLOCK_INITIALIZER; + * This is a recursive mutex to allow recursive acquisition when calling into + * providers. For example, netdev_run() calls into provider 'run' functions, + * which might reasonably want to call one of the netdev functions that takes + * netdev_class_mutex. */ +static struct ovs_mutex netdev_class_mutex OVS_ACQ_BEFORE(netdev_mutex); /* Contains 'struct netdev_registered_class'es. */ -static struct hmap netdev_classes OVS_GUARDED_BY(netdev_class_rwlock) +static struct hmap netdev_classes OVS_GUARDED_BY(netdev_class_mutex) = HMAP_INITIALIZER(&netdev_classes); struct netdev_registered_class { @@ -91,17 +91,31 @@ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); static void restore_all_flags(void *aux OVS_UNUSED); void update_device_args(struct netdev *, const struct shash *args); +int +netdev_n_rxq(const struct netdev *netdev) +{ + return netdev->n_rxq; +} + +bool +netdev_is_pmd(const struct netdev *netdev) +{ + return !strcmp(netdev->netdev_class->type, "dpdk"); +} + static void netdev_initialize(void) - OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; if (ovsthread_once_start(&once)) { + ovs_mutex_init_recursive(&netdev_class_mutex); + fatal_signal_add_hook(restore_all_flags, NULL, NULL, true); netdev_vport_patch_register(); -#ifdef LINUX_DATAPATH +#ifdef __linux__ netdev_register_provider(&netdev_linux_class); netdev_register_provider(&netdev_internal_class); netdev_register_provider(&netdev_tap_class); @@ -113,6 +127,7 @@ netdev_initialize(void) #endif netdev_register_provider(&netdev_tunnel_class); netdev_register_provider(&netdev_pltap_class); + netdev_dpdk_register(); ovsthread_once_done(&once); } @@ -124,17 +139,17 @@ netdev_initialize(void) * main poll loop. */ void netdev_run(void) - OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { struct netdev_registered_class *rc; - ovs_rwlock_rdlock(&netdev_class_rwlock); + ovs_mutex_lock(&netdev_class_mutex); HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { if (rc->class->run) { rc->class->run(); } } - ovs_rwlock_unlock(&netdev_class_rwlock); + ovs_mutex_unlock(&netdev_class_mutex); } /* Arranges for poll_block() to wake up when netdev_run() needs to be called. @@ -143,22 +158,22 @@ netdev_run(void) * main poll loop. */ void netdev_wait(void) - OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { struct netdev_registered_class *rc; - ovs_rwlock_rdlock(&netdev_class_rwlock); + ovs_mutex_lock(&netdev_class_mutex); HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { if (rc->class->wait) { rc->class->wait(); } } - ovs_rwlock_unlock(&netdev_class_rwlock); + ovs_mutex_unlock(&netdev_class_mutex); } static struct netdev_registered_class * netdev_lookup_class(const char *type) - OVS_REQ_RDLOCK(netdev_class_rwlock) + OVS_REQ_RDLOCK(netdev_class_mutex) { struct netdev_registered_class *rc; @@ -175,11 +190,11 @@ netdev_lookup_class(const char *type) * registration, new netdevs of that type can be opened using netdev_open(). */ int netdev_register_provider(const struct netdev_class *new_class) - OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { int error; - ovs_rwlock_wrlock(&netdev_class_rwlock); + ovs_mutex_lock(&netdev_class_mutex); if (netdev_lookup_class(new_class->type)) { VLOG_WARN("attempted to register duplicate netdev provider: %s", new_class->type); @@ -199,7 +214,7 @@ netdev_register_provider(const struct netdev_class *new_class) new_class->type, ovs_strerror(error)); } } - ovs_rwlock_unlock(&netdev_class_rwlock); + ovs_mutex_unlock(&netdev_class_mutex); return error; } @@ -209,12 +224,12 @@ netdev_register_provider(const struct netdev_class *new_class) * new netdevs of that type cannot be opened using netdev_open(). */ int netdev_unregister_provider(const char *type) - OVS_EXCLUDED(netdev_class_rwlock, netdev_mutex) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { struct netdev_registered_class *rc; int error; - ovs_rwlock_wrlock(&netdev_class_rwlock); + ovs_mutex_lock(&netdev_class_mutex); rc = netdev_lookup_class(type); if (!rc) { VLOG_WARN("attempted to unregister a netdev provider that is not " @@ -226,7 +241,6 @@ netdev_unregister_provider(const char *type) atomic_read(&rc->ref_cnt, &ref_cnt); if (!ref_cnt) { hmap_remove(&netdev_classes, &rc->hmap_node); - atomic_destroy(&rc->ref_cnt); free(rc); error = 0; } else { @@ -235,7 +249,7 @@ netdev_unregister_provider(const char *type) error = EBUSY; } } - ovs_rwlock_unlock(&netdev_class_rwlock); + ovs_mutex_unlock(&netdev_class_mutex); return error; } @@ -251,11 +265,11 @@ netdev_enumerate_types(struct sset *types) netdev_initialize(); sset_clear(types); - ovs_rwlock_rdlock(&netdev_class_rwlock); + ovs_mutex_lock(&netdev_class_mutex); HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { sset_add(types, rc->class->type); } - ovs_rwlock_unlock(&netdev_class_rwlock); + ovs_mutex_unlock(&netdev_class_mutex); } /* Check that the network device name is not the same as any of the registered @@ -271,15 +285,15 @@ netdev_is_reserved_name(const char *name) netdev_initialize(); - ovs_rwlock_rdlock(&netdev_class_rwlock); + ovs_mutex_lock(&netdev_class_mutex); HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { const char *dpif_port = netdev_vport_class_get_dpif_port(rc->class); if (dpif_port && !strcmp(dpif_port, name)) { - ovs_rwlock_unlock(&netdev_class_rwlock); + ovs_mutex_unlock(&netdev_class_mutex); return true; } } - ovs_rwlock_unlock(&netdev_class_rwlock); + ovs_mutex_unlock(&netdev_class_mutex); if (!strncmp(name, "ovs-", 4)) { struct sset types; @@ -315,7 +329,7 @@ netdev_open(const char *name, const char *type, struct netdev **netdevp) netdev_initialize(); - ovs_rwlock_rdlock(&netdev_class_rwlock); + ovs_mutex_lock(&netdev_class_mutex); ovs_mutex_lock(&netdev_mutex); netdev = shash_find_data(&netdev_shash, name); if (!netdev) { @@ -329,6 +343,13 @@ netdev_open(const char *name, const char *type, struct netdev **netdevp) netdev->netdev_class = rc->class; netdev->name = xstrdup(name); netdev->node = shash_add(&netdev_shash, name, netdev); + + /* By default enable one rx queue per netdev. */ + if (netdev->netdev_class->rxq_alloc) { + netdev->n_rxq = 1; + } else { + netdev->n_rxq = 0; + } list_init(&netdev->saved_flags_list); error = rc->class->construct(netdev); @@ -356,7 +377,7 @@ netdev_open(const char *name, const char *type, struct netdev **netdevp) } ovs_mutex_unlock(&netdev_mutex); - ovs_rwlock_unlock(&netdev_class_rwlock); + ovs_mutex_unlock(&netdev_class_mutex); if (!error) { netdev->ref_cnt++; @@ -462,11 +483,11 @@ netdev_unref(struct netdev *dev) dev->netdev_class->dealloc(dev); ovs_mutex_unlock(&netdev_mutex); - ovs_rwlock_rdlock(&netdev_class_rwlock); + ovs_mutex_lock(&netdev_class_mutex); rc = netdev_lookup_class(class->type); atomic_sub(&rc->ref_cnt, 1, &old_ref_cnt); ovs_assert(old_ref_cnt > 0); - ovs_rwlock_unlock(&netdev_class_rwlock); + ovs_mutex_unlock(&netdev_class_mutex); } else { ovs_mutex_unlock(&netdev_mutex); } @@ -502,24 +523,25 @@ netdev_parse_name(const char *netdev_name_, char **name, char **type) } } -/* Attempts to open a netdev_rx handle for obtaining packets received on - * 'netdev'. On success, returns 0 and stores a nonnull 'netdev_rx *' into +/* Attempts to open a netdev_rxq handle for obtaining packets received on + * 'netdev'. On success, returns 0 and stores a nonnull 'netdev_rxq *' into * '*rxp'. On failure, returns a positive errno value and stores NULL into * '*rxp'. * * Some kinds of network devices might not support receiving packets. This * function returns EOPNOTSUPP in that case.*/ int -netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp) +netdev_rxq_open(struct netdev *netdev, struct netdev_rxq **rxp, int id) OVS_EXCLUDED(netdev_mutex) { int error; - if (netdev->netdev_class->rx_alloc) { - struct netdev_rx *rx = netdev->netdev_class->rx_alloc(); + if (netdev->netdev_class->rxq_alloc && id < netdev->n_rxq) { + struct netdev_rxq *rx = netdev->netdev_class->rxq_alloc(); if (rx) { rx->netdev = netdev; - error = netdev->netdev_class->rx_construct(rx); + rx->queue_id = id; + error = netdev->netdev_class->rxq_construct(rx); if (!error) { ovs_mutex_lock(&netdev_mutex); netdev->ref_cnt++; @@ -528,7 +550,7 @@ netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp) *rxp = rx; return 0; } - netdev->netdev_class->rx_dealloc(rx); + netdev->netdev_class->rxq_dealloc(rx); } else { error = ENOMEM; } @@ -542,33 +564,24 @@ netdev_rx_open(struct netdev *netdev, struct netdev_rx **rxp) /* Closes 'rx'. */ void -netdev_rx_close(struct netdev_rx *rx) +netdev_rxq_close(struct netdev_rxq *rx) OVS_EXCLUDED(netdev_mutex) { if (rx) { struct netdev *netdev = rx->netdev; - netdev->netdev_class->rx_destruct(rx); - netdev->netdev_class->rx_dealloc(rx); + netdev->netdev_class->rxq_destruct(rx); + netdev->netdev_class->rxq_dealloc(rx); netdev_close(netdev); } } -/* Attempts to receive a packet from 'rx' into the tailroom of 'buffer', which - * must initially be empty. If successful, returns 0 and increments - * 'buffer->size' by the number of bytes in the received packet, otherwise a - * positive errno value. +/* Attempts to receive batch of packets from 'rx'. * * Returns EAGAIN immediately if no packet is ready to be received. * * Returns EMSGSIZE, and discards the packet, if the received packet is longer * than 'ofpbuf_tailroom(buffer)'. * - * Implementations may make use of VLAN_HEADER_LEN bytes of tailroom to - * add a VLAN header which is obtained out-of-band to the packet. If - * this occurs then VLAN_HEADER_LEN bytes of tailroom will no longer be - * available for the packet, otherwise it may be used for the packet - * itself. - * * It is advised that the tailroom of 'buffer' should be * VLAN_HEADER_LEN bytes longer than the MTU to allow space for an * out-of-band VLAN header to be added to the packet. At the very least, @@ -577,39 +590,31 @@ netdev_rx_close(struct netdev_rx *rx) * This function may be set to null if it would always return EOPNOTSUPP * anyhow. */ int -netdev_rx_recv(struct netdev_rx *rx, struct ofpbuf *buffer) +netdev_rxq_recv(struct netdev_rxq *rx, struct ofpbuf **buffers, int *cnt) { int retval; - ovs_assert(buffer->size == 0); - ovs_assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN); - - retval = rx->netdev->netdev_class->rx_recv(rx, buffer); + retval = rx->netdev->netdev_class->rxq_recv(rx, buffers, cnt); if (!retval) { COVERAGE_INC(netdev_received); - if (buffer->size < ETH_TOTAL_MIN) { - ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size); - } - return 0; - } else { - return retval; } + return retval; } /* Arranges for poll_block() to wake up when a packet is ready to be received * on 'rx'. */ void -netdev_rx_wait(struct netdev_rx *rx) +netdev_rxq_wait(struct netdev_rxq *rx) { - rx->netdev->netdev_class->rx_wait(rx); + rx->netdev->netdev_class->rxq_wait(rx); } /* Discards any packets ready to be received on 'rx'. */ int -netdev_rx_drain(struct netdev_rx *rx) +netdev_rxq_drain(struct netdev_rxq *rx) { - return (rx->netdev->netdev_class->rx_drain - ? rx->netdev->netdev_class->rx_drain(rx) + return (rx->netdev->netdev_class->rxq_drain + ? rx->netdev->netdev_class->rxq_drain(rx) : 0); } @@ -618,7 +623,7 @@ netdev_rx_drain(struct netdev_rx *rx) * immediately. Returns EMSGSIZE if a partial packet was transmitted or if * the packet is too big or too small to transmit on the device. * - * The caller retains ownership of 'buffer' in all cases. + * To retain ownership of 'buffer' caller can set may_steal to false. * * The kernel maintains a packet transmission queue, so the caller is not * expected to do additional queuing of packets. @@ -626,12 +631,12 @@ netdev_rx_drain(struct netdev_rx *rx) * Some network devices may not implement support for this function. In such * cases this function will always return EOPNOTSUPP. */ int -netdev_send(struct netdev *netdev, const struct ofpbuf *buffer) +netdev_send(struct netdev *netdev, struct ofpbuf *buffer, bool may_steal) { int error; error = (netdev->netdev_class->send - ? netdev->netdev_class->send(netdev, buffer->data, buffer->size) + ? netdev->netdev_class->send(netdev, buffer, may_steal) : EOPNOTSUPP); if (!error) { COVERAGE_INC(netdev_sent); @@ -1607,16 +1612,16 @@ netdev_get_type_from_name(const char *name) } struct netdev * -netdev_rx_get_netdev(const struct netdev_rx *rx) +netdev_rxq_get_netdev(const struct netdev_rxq *rx) { ovs_assert(rx->netdev->ref_cnt > 0); return rx->netdev; } const char * -netdev_rx_get_name(const struct netdev_rx *rx) +netdev_rxq_get_name(const struct netdev_rxq *rx) { - return netdev_get_name(netdev_rx_get_netdev(rx)); + return netdev_get_name(netdev_rxq_get_netdev(rx)); } static void