[ofa-general] [PATCH] ipoib - scatter/gather suopport
Eli Cohen
eli at mellanox.co.il
Sun Aug 12 04:43:33 PDT 2007
Add scatter/gather support to ipoib
Signed-off-by: Eli Cohen <eli at mellnaox.co.il>
---
Index: linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_main.c
===================================================================
--- linux-2.6.23-rc1.orig/drivers/infiniband/ulp/ipoib/ipoib_main.c 2007-08-02 14:08:29.000000000 +0300
+++ linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_main.c 2007-08-02 14:31:02.000000000 +0300
@@ -1083,7 +1083,7 @@ static struct net_device *ipoib_add_port
pdev = to_pci_dev(hca->dma_device);
if (pdev->dma_mask & DMA_64BIT_MASK)
- priv->dev->features |= NETIF_F_HIGHDMA;
+ priv->dev->features |= NETIF_F_HIGHDMA | NETIF_F_SG;
result = ib_query_pkey(hca, port, 0, &priv->pkey);
if (result) {
Index: linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib.h
===================================================================
--- linux-2.6.23-rc1.orig/drivers/infiniband/ulp/ipoib/ipoib.h 2007-07-10 21:56:30.000000000 +0300
+++ linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib.h 2007-08-02 14:31:02.000000000 +0300
@@ -120,9 +120,14 @@ struct ipoib_rx_buf {
u64 mapping;
};
+struct ipoib_mapping_st {
+ u64 addr;
+ u16 size;
+};
+
struct ipoib_tx_buf {
struct sk_buff *skb;
- u64 mapping;
+ struct ipoib_mapping_st mapping[MAX_SKB_FRAGS + 1];
};
struct ib_cm_id;
@@ -269,7 +274,7 @@ struct ipoib_dev_priv {
struct ipoib_tx_buf *tx_ring;
unsigned tx_head;
unsigned tx_tail;
- struct ib_sge tx_sge;
+ struct ib_sge tx_sge[MAX_SKB_FRAGS + 1];
struct ib_send_wr tx_wr;
struct ib_wc ibwc[IPOIB_NUM_WC];
Index: linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_ib.c
===================================================================
--- linux-2.6.23-rc1.orig/drivers/infiniband/ulp/ipoib/ipoib_ib.c 2007-08-02 11:44:43.000000000 +0300
+++ linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_ib.c 2007-08-02 14:31:02.000000000 +0300
@@ -239,6 +239,24 @@ repost:
"for buf %d\n", wr_id);
}
+static int dma_unmap_list(struct ib_device *ca, struct ipoib_mapping_st *map,
+ u16 n)
+{
+ int i, len;
+
+ BUG_ON(!n);
+ ib_dma_unmap_single(ca, map[0].addr, map[0].size, DMA_TO_DEVICE);
+ len = map[0].size;
+
+ for (i = 1; i < n; ++i) {
+ ib_dma_unmap_page(ca, map[i].addr, map[i].size,
+ DMA_TO_DEVICE);
+ len += map[i].size;
+ }
+
+ return len;
+}
+
static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
{
struct ipoib_dev_priv *priv = netdev_priv(dev);
@@ -256,12 +274,9 @@ static void ipoib_ib_handle_tx_wc(struct
}
tx_req = &priv->tx_ring[wr_id];
-
- ib_dma_unmap_single(priv->ca, tx_req->mapping,
- tx_req->skb->len, DMA_TO_DEVICE);
-
+ priv->stats.tx_bytes += dma_unmap_list(priv->ca, tx_req->mapping,
+ skb_shinfo(tx_req->skb)->nr_frags + 1);
++priv->stats.tx_packets;
- priv->stats.tx_bytes += tx_req->skb->len;
dev_kfree_skb_any(tx_req->skb);
@@ -343,13 +358,17 @@ void ipoib_ib_completion(struct ib_cq *c
static inline int post_send(struct ipoib_dev_priv *priv,
unsigned int wr_id,
struct ib_ah *address, u32 qpn,
- u64 addr, int len)
+ struct ipoib_mapping_st *mapping, int ngather)
{
struct ib_send_wr *bad_wr;
+ int i;
- priv->tx_sge.addr = addr;
- priv->tx_sge.length = len;
+ for (i = 0; i < ngather; ++i) {
+ priv->tx_sge[i].addr = mapping[i].addr;
+ priv->tx_sge[i].length = mapping[i].size;
+ }
+ priv->tx_wr.num_sge = ngather;
priv->tx_wr.wr_id = wr_id;
priv->tx_wr.wr.ud.remote_qpn = qpn;
priv->tx_wr.wr.ud.ah = address;
@@ -362,7 +381,9 @@ void ipoib_send(struct net_device *dev,
{
struct ipoib_dev_priv *priv = netdev_priv(dev);
struct ipoib_tx_buf *tx_req;
+ struct skb_frag_struct *frag;
u64 addr;
+ unsigned short i;
if (unlikely(skb->len > priv->mcast_mtu + IPOIB_ENCAP_LEN)) {
ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n",
@@ -385,21 +406,32 @@ void ipoib_send(struct net_device *dev,
*/
tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)];
tx_req->skb = skb;
- addr = ib_dma_map_single(priv->ca, skb->data, skb->len,
+
+ addr = ib_dma_map_single(priv->ca, skb->data, skb_headlen(skb),
DMA_TO_DEVICE);
if (unlikely(ib_dma_mapping_error(priv->ca, addr))) {
++priv->stats.tx_errors;
dev_kfree_skb_any(skb);
return;
}
- tx_req->mapping = addr;
+ tx_req->mapping[0].addr = addr;
+ tx_req->mapping[0].size = skb_headlen(skb);
+
+ for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) {
+ frag = &skb_shinfo(skb)->frags[i];
+ addr = ib_dma_map_page(priv->ca, frag->page, frag->page_offset,
+ frag->size, DMA_TO_DEVICE);
+ if (unlikely(ib_dma_mapping_error(priv->ca, addr)))
+ goto map_err;
+
+ tx_req->mapping[i + 1].addr = addr;
+ tx_req->mapping[i + 1].size = frag->size;
+ }
if (unlikely(post_send(priv, priv->tx_head & (ipoib_sendq_size - 1),
- address->ah, qpn, addr, skb->len))) {
+ address->ah, qpn, tx_req->mapping, skb_shinfo(skb)->nr_frags + 1))) {
ipoib_warn(priv, "post_send failed\n");
- ++priv->stats.tx_errors;
- ib_dma_unmap_single(priv->ca, addr, skb->len, DMA_TO_DEVICE);
- dev_kfree_skb_any(skb);
+ goto map_err;
} else {
dev->trans_start = jiffies;
@@ -412,6 +444,11 @@ void ipoib_send(struct net_device *dev,
set_bit(IPOIB_FLAG_NETIF_STOPPED, &priv->flags);
}
}
+ return;
+
+map_err:
+ dma_unmap_list(priv->ca, tx_req->mapping, i + 1);
+ dev_kfree_skb_any(skb);
}
static void __ipoib_reap_ah(struct net_device *dev)
@@ -604,10 +641,8 @@ int ipoib_ib_dev_stop(struct net_device
while ((int) priv->tx_tail - (int) priv->tx_head < 0) {
tx_req = &priv->tx_ring[priv->tx_tail &
(ipoib_sendq_size - 1)];
- ib_dma_unmap_single(priv->ca,
- tx_req->mapping,
- tx_req->skb->len,
- DMA_TO_DEVICE);
+ dma_unmap_list(priv->ca, tx_req->mapping,
+ skb_shinfo(tx_req->skb)->nr_frags + 1);
dev_kfree_skb_any(tx_req->skb);
++priv->tx_tail;
}
Index: linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
===================================================================
--- linux-2.6.23-rc1.orig/drivers/infiniband/ulp/ipoib/ipoib_verbs.c 2007-07-10 21:56:30.000000000 +0300
+++ linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_verbs.c 2007-08-02 14:31:02.000000000 +0300
@@ -149,14 +149,14 @@ int ipoib_transport_dev_init(struct net_
.cap = {
.max_send_wr = ipoib_sendq_size,
.max_recv_wr = ipoib_recvq_size,
- .max_send_sge = 1,
+ .max_send_sge = MAX_SKB_FRAGS + 1,
.max_recv_sge = 1
},
.sq_sig_type = IB_SIGNAL_ALL_WR,
.qp_type = IB_QPT_UD
};
- int ret, size;
+ int ret, size, i;
priv->pd = ib_alloc_pd(priv->ca);
if (IS_ERR(priv->pd)) {
@@ -197,11 +197,11 @@ int ipoib_transport_dev_init(struct net_
priv->dev->dev_addr[2] = (priv->qp->qp_num >> 8) & 0xff;
priv->dev->dev_addr[3] = (priv->qp->qp_num ) & 0xff;
- priv->tx_sge.lkey = priv->mr->lkey;
+ for (i = 0; i < MAX_SKB_FRAGS + 1; ++i)
+ priv->tx_sge[i].lkey = priv->mr->lkey;
priv->tx_wr.opcode = IB_WR_SEND;
- priv->tx_wr.sg_list = &priv->tx_sge;
- priv->tx_wr.num_sge = 1;
+ priv->tx_wr.sg_list = priv->tx_sge;
priv->tx_wr.send_flags = IB_SEND_SIGNALED;
return 0;
Index: linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_cm.c
===================================================================
--- linux-2.6.23-rc1.orig/drivers/infiniband/ulp/ipoib/ipoib_cm.c 2007-08-02 11:44:43.000000000 +0300
+++ linux-2.6.23-rc1/drivers/infiniband/ulp/ipoib/ipoib_cm.c 2007-08-02 14:31:02.000000000 +0300
@@ -495,8 +495,8 @@ static inline int post_send(struct ipoib
{
struct ib_send_wr *bad_wr;
- priv->tx_sge.addr = addr;
- priv->tx_sge.length = len;
+ priv->tx_sge[0].addr = addr;
+ priv->tx_sge[0].length = len;
priv->tx_wr.wr_id = wr_id;
@@ -537,7 +537,7 @@ void ipoib_cm_send(struct net_device *de
return;
}
- tx_req->mapping = addr;
+ tx_req->mapping[0].addr = addr;
if (unlikely(post_send(priv, tx, tx->tx_head & (ipoib_sendq_size - 1),
addr, skb->len))) {
@@ -577,7 +577,7 @@ static void ipoib_cm_handle_tx_wc(struct
tx_req = &tx->tx_ring[wr_id];
- ib_dma_unmap_single(priv->ca, tx_req->mapping, tx_req->skb->len, DMA_TO_DEVICE);
+ ib_dma_unmap_single(priv->ca, tx_req->mapping[0].addr, tx_req->skb->len, DMA_TO_DEVICE);
/* FIXME: is this right? Shouldn't we only increment on success? */
++priv->stats.tx_packets;
@@ -981,7 +981,7 @@ static void ipoib_cm_tx_destroy(struct i
if (p->tx_ring) {
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, tx_req->skb->len,
+ ib_dma_unmap_single(priv->ca, tx_req->mapping[0].addr, tx_req->skb->len,
DMA_TO_DEVICE);
dev_kfree_skb_any(tx_req->skb);
++p->tx_tail;
More information about the general
mailing list