diff options
author | David S. Miller <davem@davemloft.net> | 2008-07-18 02:39:39 -0700 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2008-07-18 02:39:39 -0700 |
commit | 49997d75152b3d23c53b0fa730599f2f74c92c65 (patch) | |
tree | 46e93126170d02cfec9505172e545732c1b69656 /drivers/infiniband/ulp/ipoib | |
parent | a0c80b80e0fb48129e4e9d6a9ede914f9ff1850d (diff) | |
parent | 5b664cb235e97afbf34db9c4d77f08ebd725335e (diff) |
Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/torvalds/linux-2.6
Conflicts:
Documentation/powerpc/booting-without-of.txt
drivers/atm/Makefile
drivers/net/fs_enet/fs_enet-main.c
drivers/pci/pci-acpi.c
net/8021q/vlan.c
net/iucv/iucv.c
Diffstat (limited to 'drivers/infiniband/ulp/ipoib')
-rw-r--r-- | drivers/infiniband/ulp/ipoib/Kconfig | 1 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib.h | 48 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_cm.c | 104 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_ethtool.c | 46 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_fs.c | 2 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_ib.c | 52 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_main.c | 115 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_multicast.c | 27 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_verbs.c | 69 | ||||
-rw-r--r-- | drivers/infiniband/ulp/ipoib/ipoib_vlan.c | 2 |
10 files changed, 342 insertions, 124 deletions
diff --git a/drivers/infiniband/ulp/ipoib/Kconfig b/drivers/infiniband/ulp/ipoib/Kconfig index 1f76bad020f3..691525cf394a 100644 --- a/drivers/infiniband/ulp/ipoib/Kconfig +++ b/drivers/infiniband/ulp/ipoib/Kconfig @@ -1,6 +1,7 @@ config INFINIBAND_IPOIB tristate "IP-over-InfiniBand" depends on NETDEVICES && INET && (IPV6 || IPV6=n) + select INET_LRO ---help--- Support for the IP-over-InfiniBand protocol (IPoIB). This transports IP packets over InfiniBand so you can use your IB diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h index ca126fc2b853..b0ffc9abe8c0 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib.h +++ b/drivers/infiniband/ulp/ipoib/ipoib.h @@ -30,8 +30,6 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * - * $Id: ipoib.h 1358 2004-12-17 22:00:11Z roland $ */ #ifndef _IPOIB_H @@ -52,9 +50,16 @@ #include <rdma/ib_verbs.h> #include <rdma/ib_pack.h> #include <rdma/ib_sa.h> +#include <linux/inet_lro.h> /* constants */ +enum ipoib_flush_level { + IPOIB_FLUSH_LIGHT, + IPOIB_FLUSH_NORMAL, + IPOIB_FLUSH_HEAVY +}; + enum { IPOIB_ENCAP_LEN = 4, @@ -65,8 +70,8 @@ enum { IPOIB_CM_BUF_SIZE = IPOIB_CM_MTU + IPOIB_ENCAP_LEN, IPOIB_CM_HEAD_SIZE = IPOIB_CM_BUF_SIZE % PAGE_SIZE, IPOIB_CM_RX_SG = ALIGN(IPOIB_CM_BUF_SIZE, PAGE_SIZE) / PAGE_SIZE, - IPOIB_RX_RING_SIZE = 128, - IPOIB_TX_RING_SIZE = 64, + IPOIB_RX_RING_SIZE = 256, + IPOIB_TX_RING_SIZE = 128, IPOIB_MAX_QUEUE_SIZE = 8192, IPOIB_MIN_QUEUE_SIZE = 2, IPOIB_CM_MAX_CONN_QP = 4096, @@ -84,7 +89,6 @@ enum { IPOIB_FLAG_SUBINTERFACE = 5, IPOIB_MCAST_RUN = 6, IPOIB_STOP_REAPER = 7, - IPOIB_MCAST_STARTED = 8, IPOIB_FLAG_ADMIN_CM = 9, IPOIB_FLAG_UMCAST = 10, IPOIB_FLAG_CSUM = 11, @@ -96,7 +100,11 @@ enum { IPOIB_MCAST_FLAG_BUSY = 2, /* joining or already joined */ IPOIB_MCAST_FLAG_ATTACHED = 3, + IPOIB_MAX_LRO_DESCRIPTORS = 8, + IPOIB_LRO_MAX_AGGR = 64, + MAX_SEND_CQE = 16, + IPOIB_CM_COPYBREAK = 256, }; #define IPOIB_OP_RECV (1ul << 31) @@ -149,6 +157,11 @@ struct ipoib_tx_buf { u64 mapping[MAX_SKB_FRAGS + 1]; }; +struct ipoib_cm_tx_buf { + struct sk_buff *skb; + u64 mapping; +}; + struct ib_cm_id; struct ipoib_cm_data { @@ -207,7 +220,7 @@ struct ipoib_cm_tx { struct net_device *dev; struct ipoib_neigh *neigh; struct ipoib_path *path; - struct ipoib_tx_buf *tx_ring; + struct ipoib_cm_tx_buf *tx_ring; unsigned tx_head; unsigned tx_tail; unsigned long flags; @@ -249,6 +262,11 @@ struct ipoib_ethtool_st { u16 max_coalesced_frames; }; +struct ipoib_lro { + struct net_lro_mgr lro_mgr; + struct net_lro_desc lro_desc[IPOIB_MAX_LRO_DESCRIPTORS]; +}; + /* * Device private locking: tx_lock protects members used in TX fast * path (and we use LLTX so upper layers don't do extra locking). @@ -264,7 +282,6 @@ struct ipoib_dev_priv { unsigned long flags; - struct mutex mcast_mutex; struct mutex vlan_mutex; struct rb_root path_tree; @@ -276,10 +293,11 @@ struct ipoib_dev_priv { struct delayed_work pkey_poll_task; struct delayed_work mcast_task; - struct work_struct flush_task; + struct work_struct flush_light; + struct work_struct flush_normal; + struct work_struct flush_heavy; struct work_struct restart_task; struct delayed_work ah_reap_task; - struct work_struct pkey_event_task; struct ib_device *ca; u8 port; @@ -335,6 +353,8 @@ struct ipoib_dev_priv { int hca_caps; struct ipoib_ethtool_st ethtool; struct timer_list poll_timer; + + struct ipoib_lro lro; }; struct ipoib_ah { @@ -359,6 +379,7 @@ struct ipoib_path { struct rb_node rb_node; struct list_head list; + int valid; }; struct ipoib_neigh { @@ -423,11 +444,14 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_ah *address, u32 qpn); void ipoib_reap_ah(struct work_struct *work); +void ipoib_mark_paths_invalid(struct net_device *dev); void ipoib_flush_paths(struct net_device *dev); struct ipoib_dev_priv *ipoib_intf_alloc(const char *format); int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port); -void ipoib_ib_dev_flush(struct work_struct *work); +void ipoib_ib_dev_flush_light(struct work_struct *work); +void ipoib_ib_dev_flush_normal(struct work_struct *work); +void ipoib_ib_dev_flush_heavy(struct work_struct *work); void ipoib_pkey_event(struct work_struct *work); void ipoib_ib_dev_cleanup(struct net_device *dev); @@ -466,9 +490,7 @@ void ipoib_path_iter_read(struct ipoib_path_iter *iter, #endif int ipoib_mcast_attach(struct net_device *dev, u16 mlid, - union ib_gid *mgid); -int ipoib_mcast_detach(struct net_device *dev, u16 mlid, - union ib_gid *mgid); + union ib_gid *mgid, int set_qkey); int ipoib_init_qp(struct net_device *dev); int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca); diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c index 97e67d36378f..0f2d3045061a 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c @@ -28,8 +28,6 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * - * $Id$ */ #include <rdma/ib_cm.h> @@ -113,18 +111,20 @@ static int ipoib_cm_post_receive_srq(struct net_device *dev, int id) } static int ipoib_cm_post_receive_nonsrq(struct net_device *dev, - struct ipoib_cm_rx *rx, int id) + struct ipoib_cm_rx *rx, + struct ib_recv_wr *wr, + struct ib_sge *sge, int id) { struct ipoib_dev_priv *priv = netdev_priv(dev); struct ib_recv_wr *bad_wr; int i, ret; - priv->cm.rx_wr.wr_id = id | IPOIB_OP_CM | IPOIB_OP_RECV; + wr->wr_id = id | IPOIB_OP_CM | IPOIB_OP_RECV; for (i = 0; i < IPOIB_CM_RX_SG; ++i) - priv->cm.rx_sge[i].addr = rx->rx_ring[id].mapping[i]; + sge[i].addr = rx->rx_ring[id].mapping[i]; - ret = ib_post_recv(rx->qp, &priv->cm.rx_wr, &bad_wr); + ret = ib_post_recv(rx->qp, wr, &bad_wr); if (unlikely(ret)) { ipoib_warn(priv, "post recv failed for buf %d (%d)\n", id, ret); ipoib_cm_dma_unmap_rx(priv, IPOIB_CM_RX_SG - 1, @@ -322,10 +322,33 @@ static int ipoib_cm_modify_rx_qp(struct net_device *dev, return 0; } +static void ipoib_cm_init_rx_wr(struct net_device *dev, + struct ib_recv_wr *wr, + struct ib_sge *sge) +{ + struct ipoib_dev_priv *priv = netdev_priv(dev); + int i; + + for (i = 0; i < priv->cm.num_frags; ++i) + sge[i].lkey = priv->mr->lkey; + + sge[0].length = IPOIB_CM_HEAD_SIZE; + for (i = 1; i < priv->cm.num_frags; ++i) + sge[i].length = PAGE_SIZE; + + wr->next = NULL; + wr->sg_list = priv->cm.rx_sge; + wr->num_sge = priv->cm.num_frags; +} + static int ipoib_cm_nonsrq_init_rx(struct net_device *dev, struct ib_cm_id *cm_id, struct ipoib_cm_rx *rx) { struct ipoib_dev_priv *priv = netdev_priv(dev); + struct { + struct ib_recv_wr wr; + struct ib_sge sge[IPOIB_CM_RX_SG]; + } *t; int ret; int i; @@ -333,6 +356,14 @@ static int ipoib_cm_nonsrq_init_rx(struct net_device *dev, struct ib_cm_id *cm_i if (!rx->rx_ring) return -ENOMEM; + t = kmalloc(sizeof *t, GFP_KERNEL); + if (!t) { + ret = -ENOMEM; + goto err_free; + } + + ipoib_cm_init_rx_wr(dev, &t->wr, t->sge); + spin_lock_irq(&priv->lock); if (priv->cm.nonsrq_conn_qp >= ipoib_max_conn_qp) { @@ -351,8 +382,8 @@ static int ipoib_cm_nonsrq_init_rx(struct net_device *dev, struct ib_cm_id *cm_i ipoib_warn(priv, "failed to allocate receive buffer %d\n", i); ret = -ENOMEM; goto err_count; - } - ret = ipoib_cm_post_receive_nonsrq(dev, rx, i); + } + ret = ipoib_cm_post_receive_nonsrq(dev, rx, &t->wr, t->sge, i); if (ret) { ipoib_warn(priv, "ipoib_cm_post_receive_nonsrq " "failed for buf %d\n", i); @@ -363,6 +394,8 @@ static int ipoib_cm_nonsrq_init_rx(struct net_device *dev, struct ib_cm_id *cm_i rx->recv_count = ipoib_recvq_size; + kfree(t); + return 0; err_count: @@ -371,6 +404,7 @@ err_count: spin_unlock_irq(&priv->lock); err_free: + kfree(t); ipoib_cm_free_rx_ring(dev, rx->rx_ring); return ret; @@ -525,6 +559,7 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) u64 mapping[IPOIB_CM_RX_SG]; int frags; int has_srq; + struct sk_buff *small_skb; ipoib_dbg_data(priv, "cm recv completion: id %d, status: %d\n", wr_id, wc->status); @@ -579,6 +614,23 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) } } + if (wc->byte_len < IPOIB_CM_COPYBREAK) { + int dlen = wc->byte_len; + + small_skb = dev_alloc_skb(dlen + 12); + if (small_skb) { + skb_reserve(small_skb, 12); + ib_dma_sync_single_for_cpu(priv->ca, rx_ring[wr_id].mapping[0], + dlen, DMA_FROM_DEVICE); + skb_copy_from_linear_data(skb, small_skb->data, dlen); + ib_dma_sync_single_for_device(priv->ca, rx_ring[wr_id].mapping[0], + dlen, DMA_FROM_DEVICE); + skb_put(small_skb, dlen); + skb = small_skb; + goto copied; + } + } + frags = PAGE_ALIGN(wc->byte_len - min(wc->byte_len, (unsigned)IPOIB_CM_HEAD_SIZE)) / PAGE_SIZE; @@ -601,6 +653,7 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) skb_put_frags(skb, IPOIB_CM_HEAD_SIZE, wc->byte_len, newskb); +copied: skb->protocol = ((struct ipoib_header *) skb->data)->proto; skb_reset_mac_header(skb); skb_pull(skb, IPOIB_ENCAP_LEN); @@ -620,7 +673,10 @@ repost: ipoib_warn(priv, "ipoib_cm_post_receive_srq failed " "for buf %d\n", wr_id); } else { - if (unlikely(ipoib_cm_post_receive_nonsrq(dev, p, wr_id))) { + if (unlikely(ipoib_cm_post_receive_nonsrq(dev, p, + &priv->cm.rx_wr, + priv->cm.rx_sge, + wr_id))) { --p->recv_count; ipoib_warn(priv, "ipoib_cm_post_receive_nonsrq failed " "for buf %d\n", wr_id); @@ -647,7 +703,7 @@ static inline int post_send(struct ipoib_dev_priv *priv, void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx) { struct ipoib_dev_priv *priv = netdev_priv(dev); - struct ipoib_tx_buf *tx_req; + struct ipoib_cm_tx_buf *tx_req; u64 addr; if (unlikely(skb->len > tx->mtu)) { @@ -678,7 +734,7 @@ void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_ return; } - tx_req->mapping[0] = addr; + tx_req->mapping = addr; if (unlikely(post_send(priv, tx, tx->tx_head & (ipoib_sendq_size - 1), addr, skb->len))) { @@ -703,7 +759,7 @@ void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) struct ipoib_dev_priv *priv = netdev_priv(dev); struct ipoib_cm_tx *tx = wc->qp->qp_context; unsigned int wr_id = wc->wr_id & ~IPOIB_OP_CM; - struct ipoib_tx_buf *tx_req; + struct ipoib_cm_tx_buf *tx_req; unsigned long flags; ipoib_dbg_data(priv, "cm send completion: id %d, status: %d\n", @@ -717,7 +773,7 @@ void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) tx_req = &tx->tx_ring[wr_id]; - ib_dma_unmap_single(priv->ca, tx_req->mapping[0], tx_req->skb->len, DMA_TO_DEVICE); + ib_dma_unmap_single(priv->ca, tx_req->mapping, tx_req->skb->len, DMA_TO_DEVICE); /* FIXME: is this right? Shouldn't we only increment on success? */ ++dev->stats.tx_packets; @@ -1087,7 +1143,7 @@ err_tx: static void ipoib_cm_tx_destroy(struct ipoib_cm_tx *p) { struct ipoib_dev_priv *priv = netdev_priv(p->dev); - struct ipoib_tx_buf *tx_req; + struct ipoib_cm_tx_buf *tx_req; unsigned long flags; unsigned long begin; @@ -1115,7 +1171,7 @@ timeout: while ((int) p->tx_tail - (int) p->tx_head < 0) { tx_req = &p->tx_ring[p->tx_tail & (ipoib_sendq_size - 1)]; - ib_dma_unmap_single(priv->ca, tx_req->mapping[0], tx_req->skb->len, + ib_dma_unmap_single(priv->ca, tx_req->mapping, tx_req->skb->len, DMA_TO_DEVICE); dev_kfree_skb_any(tx_req->skb); ++p->tx_tail; @@ -1384,7 +1440,9 @@ static ssize_t set_mode(struct device *d, struct device_attribute *attr, ipoib_warn(priv, "enabling connected mode " "will cause multicast packet drops\n"); + rtnl_lock(); dev->features &= ~(NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO); + rtnl_unlock(); priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM; ipoib_flush_paths(dev); @@ -1393,14 +1451,16 @@ static ssize_t set_mode(struct device *d, struct device_attribute *attr, if (!strcmp(buf, "datagram\n")) { clear_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags); - dev->mtu = min(priv->mcast_mtu, dev->mtu); - ipoib_flush_paths(dev); + rtnl_lock(); if (test_bit(IPOIB_FLAG_CSUM, &priv->flags)) { dev->features |= NETIF_F_IP_CSUM | NETIF_F_SG; if (priv->hca_caps & IB_DEVICE_UD_TSO) dev->features |= NETIF_F_TSO; } + dev_set_mtu(dev, min(priv->mcast_mtu, dev->mtu)); + rtnl_unlock(); + ipoib_flush_paths(dev); return count; } @@ -1485,15 +1545,7 @@ int ipoib_cm_dev_init(struct net_device *dev) priv->cm.num_frags = IPOIB_CM_RX_SG; } - for (i = 0; i < priv->cm.num_frags; ++i) - priv->cm.rx_sge[i].lkey = priv->mr->lkey; - - priv->cm.rx_sge[0].length = IPOIB_CM_HEAD_SIZE; - for (i = 1; i < priv->cm.num_frags; ++i) - priv->cm.rx_sge[i].length = PAGE_SIZE; - priv->cm.rx_wr.next = NULL; - priv->cm.rx_wr.sg_list = priv->cm.rx_sge; - priv->cm.rx_wr.num_sge = priv->cm.num_frags; + ipoib_cm_init_rx_wr(dev, &priv->cm.rx_wr, priv->cm.rx_sge); if (ipoib_cm_has_srq(dev)) { for (i = 0; i < ipoib_recvq_size; ++i) { diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c b/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c index 10279b79c44d..66af5c1a76e5 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c @@ -86,11 +86,57 @@ static int ipoib_set_coalesce(struct net_device *dev, return 0; } +static const char ipoib_stats_keys[][ETH_GSTRING_LEN] = { + "LRO aggregated", "LRO flushed", + "LRO avg aggr", "LRO no desc" +}; + +static void ipoib_get_strings(struct net_device *netdev, u32 stringset, u8 *data) +{ + switch (stringset) { + case ETH_SS_STATS: + memcpy(data, *ipoib_stats_keys, sizeof(ipoib_stats_keys)); + break; + } +} + +static int ipoib_get_sset_count(struct net_device *dev, int sset) +{ + switch (sset) { + case ETH_SS_STATS: + return ARRAY_SIZE(ipoib_stats_keys); + default: + return -EOPNOTSUPP; + } +} + +static void ipoib_get_ethtool_stats(struct net_device *dev, + struct ethtool_stats *stats, uint64_t *data) +{ + struct ipoib_dev_priv *priv = netdev_priv(dev); + int index = 0; + + /* Get LRO statistics */ + data[index++] = priv->lro.lro_mgr.stats.aggregated; + data[index++] = priv->lro.lro_mgr.stats.flushed; + if (priv->lro.lro_mgr.stats.flushed) + data[index++] = priv->lro.lro_mgr.stats.aggregated / + priv->lro.lro_mgr.stats.flushed; + else + data[index++] = 0; + data[index++] = priv->lro.lro_mgr.stats.no_desc; +} + static const struct ethtool_ops ipoib_ethtool_ops = { .get_drvinfo = ipoib_get_drvinfo, .get_tso = ethtool_op_get_tso, .get_coalesce = ipoib_get_coalesce, .set_coalesce = ipoib_set_coalesce, + .get_flags = ethtool_op_get_flags, + .set_flags = ethtool_op_set_flags, + .get_strings = ipoib_get_strings, + .get_sset_count = ipoib_get_sset_count, + .get_ethtool_stats = ipoib_get_ethtool_stats, }; void ipoib_set_ethtool_ops(struct net_device *dev) diff --git a/drivers/infiniband/ulp/ipoib/ipoib_fs.c b/drivers/infiniband/ulp/ipoib/ipoib_fs.c index 8b882bbd1d05..961c585da216 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_fs.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_fs.c @@ -28,8 +28,6 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * - * $Id: ipoib_fs.c 1389 2004-12-27 22:56:47Z roland $ */ #include <linux/err.h> diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c index f429bce24c20..66cafa20c246 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c @@ -31,8 +31,6 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * - * $Id: ipoib_ib.c 1386 2004-12-27 16:23:17Z roland $ */ #include <linux/delay.h> @@ -290,7 +288,10 @@ static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) if (test_bit(IPOIB_FLAG_CSUM, &priv->flags) && likely(wc->csum_ok)) skb->ip_summed = CHECKSUM_UNNECESSARY; - netif_receive_skb(skb); + if (dev->features & NETIF_F_LRO) + lro_receive_skb(&priv->lro.lro_mgr, skb, NULL); + else + netif_receive_skb(skb); repost: if (unlikely(ipoib_ib_post_receive(dev, wr_id))) @@ -442,6 +443,9 @@ poll_more: } if (done < budget) { + if (dev->features & NETIF_F_LRO) + lro_flush_all(&priv->lro.lro_mgr); + netif_rx_complete(dev, napi); if (unlikely(ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP | @@ -898,7 +902,8 @@ int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port) return 0; } -static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, int pkey_event) +static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, + enum ipoib_flush_level level) { struct ipoib_dev_priv *cpriv; struct net_device *dev = priv->dev; @@ -911,7 +916,7 @@ static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, int pkey_event) * the parent is down. */ list_for_each_entry(cpriv, &priv->child_intfs, list) - __ipoib_ib_dev_flush(cpriv, pkey_event); + __ipoib_ib_dev_flush(cpriv, level); mutex_unlock(&priv->vlan_mutex); @@ -925,7 +930,7 @@ static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, int pkey_event) return; } - if (pkey_event) { + if (level == IPOIB_FLUSH_HEAVY) { if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &new_index)) { clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); ipoib_ib_dev_down(dev, 0); @@ -943,11 +948,15 @@ static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, int pkey_event) priv->pkey_index = new_index; } - ipoib_dbg(priv, "flushing\n"); + if (level == IPOIB_FLUSH_LIGHT) { + ipoib_mark_paths_invalid(dev); + ipoib_mcast_dev_flush(dev); + } - ipoib_ib_dev_down(dev, 0); + if (level >= IPOIB_FLUSH_NORMAL) + ipoib_ib_dev_down(dev, 0); - if (pkey_event) { + if (level == IPOIB_FLUSH_HEAVY) { ipoib_ib_dev_stop(dev, 0); ipoib_ib_dev_open(dev); } @@ -957,27 +966,34 @@ static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, int pkey_event) * we get here, don't bring it back up if it's not configured up */ if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { - ipoib_ib_dev_up(dev); + if (level >= IPOIB_FLUSH_NORMAL) + ipoib_ib_dev_up(dev); ipoib_mcast_restart_task(&priv->restart_task); } } -void ipoib_ib_dev_flush(struct work_struct *work) +void ipoib_ib_dev_flush_light(struct work_struct *work) +{ + struct ipoib_dev_priv *priv = + container_of(work, struct ipoib_dev_priv, flush_light); + + __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_LIGHT); +} + +void ipoib_ib_dev_flush_normal(struct work_struct *work) { struct ipoib_dev_priv *priv = - container_of(work, struct ipoib_dev_priv, flush_task); + container_of(work, struct ipoib_dev_priv, flush_normal); - ipoib_dbg(priv, "Flushing %s\n", priv->dev->name); - __ipoib_ib_dev_flush(priv, 0); + __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_NORMAL); } -void ipoib_pkey_event(struct work_struct *work) +void ipoib_ib_dev_flush_heavy(struct work_struct *work) { struct ipoib_dev_priv *priv = - container_of(work, struct ipoib_dev_priv, pkey_event_task); + container_of(work, struct ipoib_dev_priv, flush_heavy); - ipoib_dbg(priv, "Flushing %s and restarting its QP\n", priv->dev->name); - __ipoib_ib_dev_flush(priv, 1); + __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_HEAVY); } void ipoib_ib_dev_cleanup(struct net_device *dev) diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c index 2442090ac8d1..8be9ea0436e6 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_main.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c @@ -30,8 +30,6 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * - * $Id: ipoib_main.c 1377 2004-12-23 19:57:12Z roland $ */ #include "ipoib.h" @@ -62,6 +60,15 @@ MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue"); module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444); MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue"); +static int lro; +module_param(lro, bool, 0444); +MODULE_PARM_DESC(lro, "Enable LRO (Large Receive Offload)"); + +static int lro_max_aggr = IPOIB_LRO_MAX_AGGR; +module_param(lro_max_aggr, int, 0644); +MODULE_PARM_DESC(lro_max_aggr, "LRO: Max packets to be aggregated " + "(default = 64)"); + #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG int ipoib_debug_level; @@ -350,6 +357,23 @@ void ipoib_path_iter_read(struct ipoib_path_iter *iter, #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */ +void ipoib_mark_paths_invalid(struct net_device *dev) +{ + struct ipoib_dev_priv *priv = netdev_priv(dev); + struct ipoib_path *path, *tp; + + spin_lock_irq(&priv->lock); + + list_for_each_entry_safe(path, tp, &priv->path_list, list) { + ipoib_dbg(priv, "mark path LID 0x%04x GID " IPOIB_GID_FMT " invalid\n", + be16_to_cpu(path->pathrec.dlid), + IPOIB_GID_ARG(path->pathrec.dgid)); + path->valid = 0; + } + + spin_unlock_irq(&priv->lock); +} + void ipoib_flush_paths(struct net_device *dev) { struct ipoib_dev_priv *priv = netdev_priv(dev); @@ -386,6 +410,7 @@ static void path_rec_completion(int status, struct net_device *dev = path->dev; struct ipoib_dev_priv *priv = netdev_priv(dev); struct ipoib_ah *ah = NULL; + struct ipoib_ah *old_ah; struct ipoib_neigh *neigh, *tn; struct sk_buff_head skqueue; struct sk_buff *skb; @@ -409,6 +434,7 @@ static void path_rec_completion(int status, spin_lock_irqsave(&priv->lock, flags); + old_ah = path->ah; path->ah = ah; if (ah) { @@ -421,6 +447,17 @@ static void path_rec_completion(int status, __skb_queue_tail(&skqueue, skb); list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) { + if (neigh->ah) { + WARN_ON(neigh->ah != old_ah); + /* + * Dropping the ah reference inside + * priv->lock is safe here, because we + * will hold one more reference from + * the original value of path->ah (ie + * old_ah). + */ + ipoib_put_ah(neigh->ah); + } kref_get(&path->ah->ref); neigh->ah = path->ah; memcpy(&neigh->dgid.raw, &path->pathrec.dgid.raw, @@ -443,6 +480,7 @@ static void path_rec_completion(int status, while ((skb = __skb_dequeue(&neigh->queue))) __skb_queue_tail(&skqueue, skb); } + path->valid = 1; } path->query = NULL; @@ -450,6 +488,9 @@ static void path_rec_completion(int status, spin_unlock_irqrestore(&priv->lock, flags); + if (old_ah) + ipoib_put_ah(old_ah); + while ((skb = __skb_dequeue(&skqueue))) { skb->dev = dev; if (dev_queue_xmit(skb)) @@ -623,8 +664,9 @@ static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev, spin_lock(&priv->lock); path = __path_find(dev, phdr->hwaddr + 4); - if (!path) { - path = path_rec_create(dev, phdr->hwaddr + 4); + if (!path || !path->valid) { + if (!path) + path = path_rec_create(dev, phdr->hwaddr + 4); if (path) { /* put pseudoheader back on for next time */ skb_push(skb, sizeof *phdr); @@ -938,6 +980,54 @@ static const struct header_ops ipoib_header_ops = { .create = ipoib_hard_header, }; +static int get_skb_hdr(struct sk_buff *skb, void **iphdr, + void **tcph, u64 *hdr_flags, void *priv) +{ + unsigned int ip_len; + struct iphdr *iph; + + if (unlikely(skb->protocol != htons(ETH_P_IP))) + return -1; + + /* + * In the future we may add an else clause that verifies the + * checksum and allows devices which do not calculate checksum + * to use LRO. + */ + if (unlikely(skb->ip_summed != CHECKSUM_UNNECESSARY)) + return -1; + + /* Check for non-TCP packet */ + skb_reset_network_header(skb); + iph = ip_hdr(skb); + if (iph->protocol != IPPROTO_TCP) + return -1; + + ip_len = ip_hdrlen(skb); + skb_set_transport_header(skb, ip_len); + *tcph = tcp_hdr(skb); + + /* check if IP header and TCP header are complete */ + if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb)) + return -1; + + *hdr_flags = LRO_IPV4 | LRO_TCP; + *iphdr = iph; + + return 0; +} + +static void ipoib_lro_setup(struct ipoib_dev_priv *priv) +{ + priv->lro.lro_mgr.max_aggr = lro_max_aggr; + priv->lro.lro_mgr.max_desc = IPOIB_MAX_LRO_DESCRIPTORS; + priv->lro.lro_mgr.lro_arr = priv->lro.lro_desc; + priv->lro.lro_mgr.get_skb_header = get_skb_hdr; + priv->lro.lro_mgr.features = LRO_F_NAPI; + priv->lro.lro_mgr.dev = priv->dev; + priv->lro.lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY; +} + static void ipoib_setup(struct net_device *dev) { struct ipoib_dev_priv *priv = netdev_priv(dev); @@ -977,10 +1067,11 @@ static void ipoib_setup(struct net_device *dev) priv->dev = dev; + ipoib_lro_setup(priv); + spin_lock_init(&priv->lock); spin_lock_init(&priv->tx_lock); - mutex_init(&priv->mcast_mutex); mutex_init(&priv->vlan_mutex); INIT_LIST_HEAD(&priv->path_list); @@ -989,9 +1080,10 @@ static void ipoib_setup(struct net_device *dev) INIT_LIST_HEAD(&priv->multicast_list); INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll); - INIT_WORK(&priv->pkey_event_task, ipoib_pkey_event); INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task); - INIT_WORK(&priv->flush_task, ipoib_ib_dev_flush); + INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light); + INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal); + INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy); INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task); INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah); } @@ -1154,6 +1246,9 @@ static struct net_device *ipoib_add_port(const char *format, priv->dev->features |= NETIF_F_SG | NETIF_F_IP_CSUM; } + if (lro) + priv->dev->features |= NETIF_F_LRO; + /* * Set the full membership bit, so that we join the right * broadcast group, etc. @@ -1304,6 +1399,12 @@ static int __init ipoib_init_module(void) ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP); #endif + /* + * When copying small received packets, we only copy from the + * linear data part of the SKB, so we rely on this condition. + */ + BUILD_BUG_ON(IPOIB_CM_COPYBREAK > IPOIB_CM_HEAD_SIZE); + ret = ipoib_register_debugfs(); if (ret) return ret; diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c index cd2fb955040f..8950e9546f4e 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c @@ -30,8 +30,6 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * - * $Id: ipoib_multicast.c 1362 2004-12-18 15:56:29Z roland $ */ #include <linux/skbuff.h> @@ -188,6 +186,7 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast, struct ipoib_dev_priv *priv = netdev_priv(dev); struct ipoib_ah *ah; int ret; + int set_qkey = 0; mcast->mcmember = *mcmember; @@ -202,6 +201,7 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast, priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey); spin_unlock_irq(&priv->lock); priv->tx_wr.wr.ud.remote_qkey = priv->qkey; + set_qkey = 1; } if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { @@ -214,7 +214,7 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast, } ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid), - &mcast->mcmember.mgid); + &mcast->mcmember.mgid, set_qkey); if (ret < 0) { ipoib_warn(priv, "couldn't attach QP to multicast group " IPOIB_GID_FMT "\n", @@ -575,8 +575,11 @@ void ipoib_mcast_join_task(struct work_struct *work) priv->mcast_mtu = IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu)); - if (!ipoib_cm_admin_enabled(dev)) - dev->mtu = min(priv->mcast_mtu, priv->admin_mtu); + if (!ipoib_cm_admin_enabled(dev)) { + rtnl_lock(); + dev_set_mtu(dev, min(priv->mcast_mtu, priv->admin_mtu)); + rtnl_unlock(); + } ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n"); @@ -594,10 +597,6 @@ int ipoib_mcast_start_thread(struct net_device *dev) queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0); mutex_unlock(&mcast_mutex); - spin_lock_irq(&priv->lock); - set_bit(IPOIB_MCAST_STARTED, &priv->flags); - spin_unlock_irq(&priv->lock); - return 0; } @@ -607,10 +606,6 @@ int ipoib_mcast_stop_thread(struct net_device *dev, int flush) ipoib_dbg_mcast(priv, "stopping multicast thread\n"); - spin_lock_irq(&priv->lock); - clear_bit(IPOIB_MCAST_STARTED, &priv->flags); - spin_unlock_irq(&priv->lock); - mutex_lock(&mcast_mutex); clear_bit(IPOIB_MCAST_RUN, &priv->flags); cancel_delayed_work(&priv->mcast_task); @@ -635,10 +630,10 @@ static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast) IPOIB_GID_ARG(mcast->mcmember.mgid)); /* Remove ourselves from the multicast group */ - ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid), - &mcast->mcmember.mgid); + ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid, + be16_to_cpu(mcast->mcmember.mlid)); if (ret) - ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret); + ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret); } return 0; diff --git a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c index 8766d29ce3b7..68325119f740 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c @@ -29,24 +29,17 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * - * $Id: ipoib_verbs.c 1349 2004-12-16 21:09:43Z roland $ */ #include "ipoib.h" -int ipoib_mcast_attach(struct net_device *dev, u16 mlid, union ib_gid *mgid) +int ipoib_mcast_attach(struct net_device *dev, u16 mlid, union ib_gid *mgid, int set_qkey) { struct ipoib_dev_priv *priv = netdev_priv(dev); - struct ib_qp_attr *qp_attr; + struct ib_qp_attr *qp_attr = NULL; int ret; u16 pkey_index; - ret = -ENOMEM; - qp_attr = kmalloc(sizeof *qp_attr, GFP_KERNEL); - if (!qp_attr) - goto out; - if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) { clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); ret = -ENXIO; @@ -54,18 +47,23 @@ int ipoib_mcast_attach(struct net_device *dev, u16 mlid, union ib_gid *mgid) } set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); - /* set correct QKey for QP */ - qp_attr->qkey = priv->qkey; - ret = ib_modify_qp(priv->qp, qp_attr, IB_QP_QKEY); - if (ret) { - ipoib_warn(priv, "failed to modify QP, ret = %d\n", ret); - goto out; + if (set_qkey) { + ret = -ENOMEM; + qp_attr = kmalloc(sizeof *qp_attr, GFP_KERNEL); + if (!qp_attr) + goto out; + + /* set correct QKey for QP */ + qp_attr->qkey = priv->qkey; + ret = ib_modify_qp(priv->qp, qp_attr, IB_QP_QKEY); + if (ret) { + ipoib_warn(priv, "failed to modify QP, ret = %d\n", ret); + goto out; + } } /* attach QP to multicast group */ - mutex_lock(&priv->mcast_mutex); ret = ib_attach_mcast(priv->qp, mgid, mlid); - mutex_unlock(&priv->mcast_mutex); if (ret) ipoib_warn(priv, "failed to attach to multicast group, ret = %d\n", ret); @@ -74,20 +72,6 @@ out: return ret; } -int ipoib_mcast_detach(struct net_device *dev, u16 mlid, union ib_gid *mgid) -{ - struct ipoib_dev_priv *priv = netdev_priv(dev); - int ret; - - mutex_lock(&priv->mcast_mutex); - ret = ib_detach_mcast(priv->qp, mgid, mlid); - mutex_unlock(&priv->mcast_mutex); - if (ret) - ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret); - - return ret; -} - int ipoib_init_qp(struct net_device *dev) { struct ipoib_dev_priv *priv = netdev_priv(dev); @@ -201,7 +185,10 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca) init_attr.recv_cq = priv->recv_cq; if (priv->hca_caps & IB_DEVICE_UD_TSO) - init_attr.create_flags = IB_QP_CREATE_IPOIB_UD_LSO; + init_attr.create_flags |= IB_QP_CREATE_IPOIB_UD_LSO; + + if (priv->hca_caps & IB_DEVICE_BLOCK_MULTICAST_LOOPBACK) + init_attr.create_flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK; if (dev->features & NETIF_F_SG) init_attr.cap.max_send_sge = MAX_SKB_FRAGS + 1; @@ -289,15 +276,17 @@ void ipoib_event(struct ib_event_handler *handler, if (record->element.port_num != priv->port) return; - if (record->event == IB_EVENT_PORT_ERR || - record->event == IB_EVENT_PORT_ACTIVE || - record->event == IB_EVENT_LID_CHANGE || - record->event == IB_EVENT_SM_CHANGE || + ipoib_dbg(priv, "Event %d on device %s port %d\n", record->event, + record->device->name, record->element.port_num); + + if (record->event == IB_EVENT_SM_CHANGE || record->event == IB_EVENT_CLIENT_REREGISTER) { - ipoib_dbg(priv, "Port state change event\n"); - queue_work(ipoib_workqueue, &priv->flush_task); + queue_work(ipoib_workqueue, &priv->flush_light); + } else if (record->event == IB_EVENT_PORT_ERR || + record->event == IB_EVENT_PORT_ACTIVE || + record->event == IB_EVENT_LID_CHANGE) { + queue_work(ipoib_workqueue, &priv->flush_normal); } else if (record->event == IB_EVENT_PKEY_CHANGE) { - ipoib_dbg(priv, "P_Key change event on port:%d\n", priv->port); - queue_work(ipoib_workqueue, &priv->pkey_event_task); + queue_work(ipoib_workqueue, &priv->flush_heavy); } } diff --git a/drivers/infiniband/ulp/ipoib/ipoib_vlan.c b/drivers/infiniband/ulp/ipoib/ipoib_vlan.c index 1cdb5cfb0ff1..b08eb56196d3 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_vlan.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_vlan.c @@ -28,8 +28,6 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * - * $Id: ipoib_vlan.c 1349 2004-12-16 21:09:43Z roland $ */ #include <linux/module.h> |