ipoib_cm.c

linux内核源码

	newskb = ipoib_cm_alloc_rx_skb(dev, wr_id, frags, mapping);
	if (unlikely(!newskb)) {
		/*
		 * If we can't allocate a new RX buffer, dump
		 * this packet and reuse the old buffer.
		 */
		ipoib_dbg(priv, "failed to allocate receive buffer %d\n", wr_id);
		++dev->stats.rx_dropped;
		goto repost;
	}

	ipoib_cm_dma_unmap_rx(priv, frags, priv->cm.srq_ring[wr_id].mapping);
	memcpy(priv->cm.srq_ring[wr_id].mapping, mapping, (frags + 1) * sizeof *mapping);

	ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
		       wc->byte_len, wc->slid);

	skb_put_frags(skb, IPOIB_CM_HEAD_SIZE, wc->byte_len, newskb);

	skb->protocol = ((struct ipoib_header *) skb->data)->proto;
	skb_reset_mac_header(skb);
	skb_pull(skb, IPOIB_ENCAP_LEN);

	dev->last_rx = jiffies;
	++dev->stats.rx_packets;
	dev->stats.rx_bytes += skb->len;

	skb->dev = dev;
	/* XXX get correct PACKET_ type here */
	skb->pkt_type = PACKET_HOST;
	netif_receive_skb(skb);

repost:
	if (unlikely(ipoib_cm_post_receive(dev, wr_id)))
		ipoib_warn(priv, "ipoib_cm_post_receive failed "
			   "for buf %d\n", wr_id);
}

static inline int post_send(struct ipoib_dev_priv *priv,
			    struct ipoib_cm_tx *tx,
			    unsigned int wr_id,
			    u64 addr, int len)
{
	struct ib_send_wr *bad_wr;

	priv->tx_sge.addr             = addr;
	priv->tx_sge.length           = len;

	priv->tx_wr.wr_id 	      = wr_id | IPOIB_OP_CM;

	return ib_post_send(tx->qp, &priv->tx_wr, &bad_wr);
}

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;
	u64 addr;

	if (unlikely(skb->len > tx->mtu)) {
		ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n",
			   skb->len, tx->mtu);
		++dev->stats.tx_dropped;
		++dev->stats.tx_errors;
		ipoib_cm_skb_too_long(dev, skb, tx->mtu - IPOIB_ENCAP_LEN);
		return;
	}

	ipoib_dbg_data(priv, "sending packet: head 0x%x length %d connection 0x%x\n",
		       tx->tx_head, skb->len, tx->qp->qp_num);

	/*
	 * We put the skb into the tx_ring _before_ we call post_send()
	 * because it's entirely possible that the completion handler will
	 * run before we execute anything after the post_send().  That
	 * means we have to make sure everything is properly recorded and
	 * our state is consistent before we call post_send().
	 */
	tx_req = &tx->tx_ring[tx->tx_head & (ipoib_sendq_size - 1)];
	tx_req->skb = skb;
	addr = ib_dma_map_single(priv->ca, skb->data, skb->len, DMA_TO_DEVICE);
	if (unlikely(ib_dma_mapping_error(priv->ca, addr))) {
		++dev->stats.tx_errors;
		dev_kfree_skb_any(skb);
		return;
	}

	tx_req->mapping = addr;

	if (unlikely(post_send(priv, tx, tx->tx_head & (ipoib_sendq_size - 1),
			        addr, skb->len))) {
		ipoib_warn(priv, "post_send failed\n");
		++dev->stats.tx_errors;
		ib_dma_unmap_single(priv->ca, addr, skb->len, DMA_TO_DEVICE);
		dev_kfree_skb_any(skb);
	} else {
		dev->trans_start = jiffies;
		++tx->tx_head;

		if (++priv->tx_outstanding == ipoib_sendq_size) {
			ipoib_dbg(priv, "TX ring 0x%x full, stopping kernel net queue\n",
				  tx->qp->qp_num);
			netif_stop_queue(dev);
		}
	}
}

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;
	unsigned long flags;

	ipoib_dbg_data(priv, "cm send completion: id %d, status: %d\n",
		       wr_id, wc->status);

	if (unlikely(wr_id >= ipoib_sendq_size)) {
		ipoib_warn(priv, "cm send completion event with wrid %d (> %d)\n",
			   wr_id, ipoib_sendq_size);
		return;
	}

	tx_req = &tx->tx_ring[wr_id];

	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;
	dev->stats.tx_bytes += tx_req->skb->len;

	dev_kfree_skb_any(tx_req->skb);

	spin_lock_irqsave(&priv->tx_lock, flags);
	++tx->tx_tail;
	if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) &&
	    netif_queue_stopped(dev) &&
	    test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
		netif_wake_queue(dev);

	if (wc->status != IB_WC_SUCCESS &&
	    wc->status != IB_WC_WR_FLUSH_ERR) {
		struct ipoib_neigh *neigh;

		ipoib_dbg(priv, "failed cm send event "
			   "(status=%d, wrid=%d vend_err %x)\n",
			   wc->status, wr_id, wc->vendor_err);

		spin_lock(&priv->lock);
		neigh = tx->neigh;

		if (neigh) {
			neigh->cm = NULL;
			list_del(&neigh->list);
			if (neigh->ah)
				ipoib_put_ah(neigh->ah);
			ipoib_neigh_free(dev, neigh);

			tx->neigh = NULL;
		}

		if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &tx->flags)) {
			list_move(&tx->list, &priv->cm.reap_list);
			queue_work(ipoib_workqueue, &priv->cm.reap_task);
		}

		clear_bit(IPOIB_FLAG_OPER_UP, &tx->flags);

		spin_unlock(&priv->lock);
	}

	spin_unlock_irqrestore(&priv->tx_lock, flags);
}

int ipoib_cm_dev_open(struct net_device *dev)
{
	struct ipoib_dev_priv *priv = netdev_priv(dev);
	int ret;

	if (!IPOIB_CM_SUPPORTED(dev->dev_addr))
		return 0;

	priv->cm.id = ib_create_cm_id(priv->ca, ipoib_cm_rx_handler, dev);
	if (IS_ERR(priv->cm.id)) {
		printk(KERN_WARNING "%s: failed to create CM ID\n", priv->ca->name);
		ret = PTR_ERR(priv->cm.id);
		goto err_cm;
	}

	ret = ib_cm_listen(priv->cm.id, cpu_to_be64(IPOIB_CM_IETF_ID | priv->qp->qp_num),
			   0, NULL);
	if (ret) {
		printk(KERN_WARNING "%s: failed to listen on ID 0x%llx\n", priv->ca->name,
		       IPOIB_CM_IETF_ID | priv->qp->qp_num);
		goto err_listen;
	}

	return 0;

err_listen:
	ib_destroy_cm_id(priv->cm.id);
err_cm:
	priv->cm.id = NULL;
	return ret;
}

void ipoib_cm_dev_stop(struct net_device *dev)
{
	struct ipoib_dev_priv *priv = netdev_priv(dev);
	struct ipoib_cm_rx *p, *n;
	unsigned long begin;
	LIST_HEAD(list);
	int ret;

	if (!IPOIB_CM_SUPPORTED(dev->dev_addr) || !priv->cm.id)
		return;

	ib_destroy_cm_id(priv->cm.id);
	priv->cm.id = NULL;

	spin_lock_irq(&priv->lock);
	while (!list_empty(&priv->cm.passive_ids)) {
		p = list_entry(priv->cm.passive_ids.next, typeof(*p), list);
		list_move(&p->list, &priv->cm.rx_error_list);
		p->state = IPOIB_CM_RX_ERROR;
		spin_unlock_irq(&priv->lock);
		ret = ib_modify_qp(p->qp, &ipoib_cm_err_attr, IB_QP_STATE);
		if (ret)
			ipoib_warn(priv, "unable to move qp to error state: %d\n", ret);
		spin_lock_irq(&priv->lock);
	}

	/* Wait for all RX to be drained */
	begin = jiffies;

	while (!list_empty(&priv->cm.rx_error_list) ||
	       !list_empty(&priv->cm.rx_flush_list) ||
	       !list_empty(&priv->cm.rx_drain_list)) {
		if (time_after(jiffies, begin + 5 * HZ)) {
			ipoib_warn(priv, "RX drain timing out\n");

			/*
			 * assume the HW is wedged and just free up everything.
			 */
			list_splice_init(&priv->cm.rx_flush_list, &list);
			list_splice_init(&priv->cm.rx_error_list, &list);
			list_splice_init(&priv->cm.rx_drain_list, &list);
			break;
		}
		spin_unlock_irq(&priv->lock);
		msleep(1);
		ipoib_drain_cq(dev);
		spin_lock_irq(&priv->lock);
	}

	list_splice_init(&priv->cm.rx_reap_list, &list);

	spin_unlock_irq(&priv->lock);

	list_for_each_entry_safe(p, n, &list, list) {
		ib_destroy_cm_id(p->id);
		ib_destroy_qp(p->qp);
		kfree(p);
	}

	cancel_delayed_work(&priv->cm.stale_task);
}

static int ipoib_cm_rep_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
{
	struct ipoib_cm_tx *p = cm_id->context;
	struct ipoib_dev_priv *priv = netdev_priv(p->dev);
	struct ipoib_cm_data *data = event->private_data;
	struct sk_buff_head skqueue;
	struct ib_qp_attr qp_attr;
	int qp_attr_mask, ret;
	struct sk_buff *skb;

	p->mtu = be32_to_cpu(data->mtu);

	if (p->mtu <= IPOIB_ENCAP_LEN) {
		ipoib_warn(priv, "Rejecting connection: mtu %d <= %d\n",
			   p->mtu, IPOIB_ENCAP_LEN);
		return -EINVAL;
	}

	qp_attr.qp_state = IB_QPS_RTR;
	ret = ib_cm_init_qp_attr(cm_id, &qp_attr, &qp_attr_mask);
	if (ret) {
		ipoib_warn(priv, "failed to init QP attr for RTR: %d\n", ret);
		return ret;
	}

	qp_attr.rq_psn = 0 /* FIXME */;
	ret = ib_modify_qp(p->qp, &qp_attr, qp_attr_mask);
	if (ret) {
		ipoib_warn(priv, "failed to modify QP to RTR: %d\n", ret);
		return ret;
	}

	qp_attr.qp_state = IB_QPS_RTS;
	ret = ib_cm_init_qp_attr(cm_id, &qp_attr, &qp_attr_mask);
	if (ret) {
		ipoib_warn(priv, "failed to init QP attr for RTS: %d\n", ret);
		return ret;
	}
	ret = ib_modify_qp(p->qp, &qp_attr, qp_attr_mask);
	if (ret) {
		ipoib_warn(priv, "failed to modify QP to RTS: %d\n", ret);
		return ret;
	}

	skb_queue_head_init(&skqueue);

	spin_lock_irq(&priv->lock);
	set_bit(IPOIB_FLAG_OPER_UP, &p->flags);
	if (p->neigh)
		while ((skb = __skb_dequeue(&p->neigh->queue)))
			__skb_queue_tail(&skqueue, skb);
	spin_unlock_irq(&priv->lock);

	while ((skb = __skb_dequeue(&skqueue))) {
		skb->dev = p->dev;
		if (dev_queue_xmit(skb))
			ipoib_warn(priv, "dev_queue_xmit failed "
				   "to requeue packet\n");
	}

	ret = ib_send_cm_rtu(cm_id, NULL, 0);
	if (ret) {
		ipoib_warn(priv, "failed to send RTU: %d\n", ret);
		return ret;
	}
	return 0;
}

static struct ib_qp *ipoib_cm_create_tx_qp(struct net_device *dev, struct ipoib_cm_tx *tx)
{
	struct ipoib_dev_priv *priv = netdev_priv(dev);
	struct ib_qp_init_attr attr = {
		.send_cq		= priv->cq,
		.recv_cq		= priv->cq,
		.srq			= priv->cm.srq,
		.cap.max_send_wr	= ipoib_sendq_size,
		.cap.max_send_sge	= 1,
		.sq_sig_type		= IB_SIGNAL_ALL_WR,
		.qp_type		= IB_QPT_RC,
		.qp_context		= tx
        };

	return ib_create_qp(priv->pd, &attr);
}

static int ipoib_cm_send_req(struct net_device *dev,
			     struct ib_cm_id *id, struct ib_qp *qp,
			     u32 qpn,
			     struct ib_sa_path_rec *pathrec)
{
	struct ipoib_dev_priv *priv = netdev_priv(dev);
	struct ipoib_cm_data data = {};
	struct ib_cm_req_param req = {};

	data.qpn = cpu_to_be32(priv->qp->qp_num);
	data.mtu = cpu_to_be32(IPOIB_CM_BUF_SIZE);

	req.primary_path 	      = pathrec;
	req.alternate_path 	      = NULL;
	req.service_id                = cpu_to_be64(IPOIB_CM_IETF_ID | qpn);
	req.qp_num 		      = qp->qp_num;
	req.qp_type 		      = qp->qp_type;
	req.private_data 	      = &data;
	req.private_data_len 	      = sizeof data;
	req.flow_control 	      = 0;

	req.starting_psn              = 0; /* FIXME */

	/*
	 * Pick some arbitrary defaults here; we could make these
	 * module parameters if anyone cared about setting them.
	 */
	req.responder_resources	      = 4;
	req.remote_cm_response_timeout = 20;
	req.local_cm_response_timeout  = 20;
	req.retry_count 	      = 0; /* RFC draft warns against retries */
	req.rnr_retry_count 	      = 0; /* RFC draft warns against retries */
	req.max_cm_retries 	      = 15;
	req.srq 	              = 1;
	return ib_send_cm_req(id, &req);
}

static int ipoib_cm_modify_tx_init(struct net_device *dev,
				  struct ib_cm_id *cm_id, struct ib_qp *qp)
{
	struct ipoib_dev_priv *priv = netdev_priv(dev);
	struct ib_qp_attr qp_attr;
	int qp_attr_mask, ret;
	ret = ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &qp_attr.pkey_index);
	if (ret) {
		ipoib_warn(priv, "pkey 0x%x not in cache: %d\n", priv->pkey, ret);
		return ret;
	}

	qp_attr.qp_state = IB_QPS_INIT;
	qp_attr.qp_access_flags = IB_ACCESS_LOCAL_WRITE;
	qp_attr.port_num = priv->port;
	qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS | IB_QP_PKEY_INDEX | IB_QP_PORT;

	ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
	if (ret) {
		ipoib_warn(priv, "failed to modify tx QP to INIT: %d\n", ret);
		return ret;
	}
	return 0;
}

static int ipoib_cm_tx_init(struct ipoib_cm_tx *p, u32 qpn,
			    struct ib_sa_path_rec *pathrec)
{
	struct ipoib_dev_priv *priv = netdev_priv(p->dev);
	int ret;

	p->tx_ring = kzalloc(ipoib_sendq_size * sizeof *p->tx_ring,
				GFP_KERNEL);
	if (!p->tx_ring) {
		ipoib_warn(priv, "failed to allocate tx ring\n");
		ret = -ENOMEM;
		goto err_tx;
	}

	p->qp = ipoib_cm_create_tx_qp(p->dev, p);
	if (IS_ERR(p->qp)) {
		ret = PTR_ERR(p->qp);
		ipoib_warn(priv, "failed to allocate tx qp: %d\n", ret);
		goto err_qp;
	}

	p->id = ib_create_cm_id(priv->ca, ipoib_cm_tx_handler, p);
	if (IS_ERR(p->id)) {
		ret = PTR_ERR(p->id);
		ipoib_warn(priv, "failed to create tx cm id: %d\n", ret);
		goto err_id;
	}

	ret = ipoib_cm_modify_tx_init(p->dev, p->id,  p->qp);
	if (ret) {
		ipoib_warn(priv, "failed to modify tx qp to rtr: %d\n", ret);
		goto err_modify;
	}

	ret = ipoib_cm_send_req(p->dev, p->id, p->qp, qpn, pathrec);
	if (ret) {