[ofa-general] [PATCH] IPoIB-UD S/G 4K MTU patch against OFED-1.3 RC2
Shirley Ma
mashirle at us.ibm.com
Wed Jan 30 22:19:05 PST 2008
Hello Vlad,
This is a patch build against OFED-1.3 RC2 on top of Pradeep's noSRQ
patch. I am not sure whether you can apply cleanly to your current
OFED-1.3 tree. If not, please let me know.
This patch is in our test bed. This patch has been tested against:
IPoIB-CM SRQ, IPoIB-UD 2K mtu, IPoIB 4K mtu, IPoIB-CM noSRQ for two
nodes. Cluster testing is going on.
Thanks
Shirley
This patch makes IPoIB-CM S/G RX more generic, and enables IPoIB UD S/G
4K MTU support.
--Signed-off-by: Shirley Ma <xma at us.ibm.com>
diff -urpN
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_cm.c
ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
---
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_cm.c
2008-01-27 14:20:17.000000000 -0600
+++
ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
2008-01-27 14:26:18.000000000 -0600
@@ -77,17 +77,6 @@ static struct ib_send_wr ipoib_cm_rx_dra
static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id,
struct ib_cm_event *event);
-static void ipoib_cm_dma_unmap_rx(struct ipoib_dev_priv *priv, int
frags,
- u64 mapping[IPOIB_CM_RX_SG])
-{
- int i;
-
- ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_CM_HEAD_SIZE,
DMA_FROM_DEVICE);
-
- for (i = 0; i < frags; ++i)
- ib_dma_unmap_single(priv->ca, mapping[i + 1], PAGE_SIZE,
DMA_FROM_DEVICE);
-}
-
static int ipoib_cm_post_receive_srq(struct net_device *dev, int id)
{
struct ipoib_dev_priv *priv = netdev_priv(dev);
@@ -102,8 +91,9 @@ static int ipoib_cm_post_receive_srq(str
ret = ib_post_srq_recv(priv->cm.srq, &priv->cm.rx_wr, &bad_wr);
if (unlikely(ret)) {
ipoib_warn(priv, "post srq failed for buf %d (%d)\n", id, ret);
- ipoib_cm_dma_unmap_rx(priv, priv->cm.num_frags - 1,
- priv->cm.srq_ring[id].mapping);
+ ipoib_dma_unmap_rx(priv, IPOIB_CM_RX_SG - 1,
+ IPOIB_CM_HEAD_SIZE,
+ priv->cm.srq_ring[id].mapping);
dev_kfree_skb_any(priv->cm.srq_ring[id].skb);
priv->cm.srq_ring[id].skb = NULL;
}
@@ -126,8 +116,9 @@ static int ipoib_cm_post_receive_nonsrq(
ret = ib_post_recv(rx->qp, &priv->cm.rx_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,
- rx->rx_ring[id].mapping);
+ ipoib_dma_unmap_rx(priv, IPOIB_CM_RX_SG - 1,
+ IPOIB_CM_HEAD_SIZE,
+ rx->rx_ring[id].mapping);
dev_kfree_skb_any(rx->rx_ring[id].skb);
rx->rx_ring[id].skb = NULL;
}
@@ -135,69 +126,16 @@ static int ipoib_cm_post_receive_nonsrq(
return ret;
}
-static struct sk_buff *ipoib_cm_alloc_rx_skb(struct net_device *dev,
- struct ipoib_cm_rx_buf *rx_ring,
- int id, int frags,
- u64 mapping[IPOIB_CM_RX_SG])
-{
- struct ipoib_dev_priv *priv = netdev_priv(dev);
- struct sk_buff *skb;
- int i;
-
- skb = dev_alloc_skb(IPOIB_CM_HEAD_SIZE + 12);
- if (unlikely(!skb))
- return NULL;
-
- /*
- * IPoIB adds a 4 byte header. So we need 12 more bytes to align the
- * IP header to a multiple of 16.
- */
- skb_reserve(skb, 12);
-
- mapping[0] = ib_dma_map_single(priv->ca, skb->data,
IPOIB_CM_HEAD_SIZE,
- DMA_FROM_DEVICE);
- if (unlikely(ib_dma_mapping_error(priv->ca, mapping[0]))) {
- dev_kfree_skb_any(skb);
- return NULL;
- }
-
- for (i = 0; i < frags; i++) {
- struct page *page = alloc_page(GFP_ATOMIC);
-
- if (!page)
- goto partial_error;
- skb_fill_page_desc(skb, i, page, 0, PAGE_SIZE);
-
- mapping[i + 1] = ib_dma_map_page(priv->ca,
skb_shinfo(skb)->frags[i].page,
- 0, PAGE_SIZE, DMA_FROM_DEVICE);
- if (unlikely(ib_dma_mapping_error(priv->ca, mapping[i + 1])))
- goto partial_error;
- }
-
- rx_ring[id].skb = skb;
- return skb;
-
-partial_error:
-
- ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_CM_HEAD_SIZE,
DMA_FROM_DEVICE);
-
- for (; i > 0; --i)
- ib_dma_unmap_single(priv->ca, mapping[i], PAGE_SIZE,
DMA_FROM_DEVICE);
-
- dev_kfree_skb_any(skb);
- return NULL;
-}
-
static void ipoib_cm_free_rx_ring(struct net_device *dev,
- struct ipoib_cm_rx_buf *rx_ring)
+ struct ipoib_rx_buf *rx_ring)
{
struct ipoib_dev_priv *priv = netdev_priv(dev);
int i;
for (i = 0; i < ipoib_recvq_size; ++i)
if (rx_ring[i].skb) {
- ipoib_cm_dma_unmap_rx(priv, IPOIB_CM_RX_SG - 1,
- rx_ring[i].mapping);
+ ipoib_dma_unmap_rx(priv, IPOIB_CM_RX_SG - 1,
+ IPOIB_CM_HEAD_SIZE, rx_ring[i].mapping);
dev_kfree_skb_any(rx_ring[i].skb);
}
@@ -345,9 +283,11 @@ static int ipoib_cm_nonsrq_init_rx(struc
spin_unlock_irq(&priv->lock);
for (i = 0; i < ipoib_recvq_size; ++i) {
- if (!ipoib_cm_alloc_rx_skb(dev, rx->rx_ring, i, IPOIB_CM_RX_SG - 1,
- rx->rx_ring[i].mapping)) {
- ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
+ rx->rx_ring[i].skb = ipoib_alloc_rx_skb(dev, i, IPOIB_CM_RX_SG - 1,
+ IPOIB_CM_HEAD_SIZE, 12,
+ rx->rx_ring[i].mapping);
+ if (!rx->rx_ring[i].skb) {
+ ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
ret = -ENOMEM;
goto err_count;
}
@@ -480,43 +420,11 @@ static int ipoib_cm_rx_handler(struct ib
return 0;
}
}
-/* Adjust length of skb with fragments to match received data */
-static void skb_put_frags(struct sk_buff *skb, unsigned int hdr_space,
- unsigned int length, struct sk_buff *toskb)
-{
- int i, num_frags;
- unsigned int size;
-
- /* put header into skb */
- size = min(length, hdr_space);
- skb->tail += size;
- skb->len += size;
- length -= size;
-
- num_frags = skb_shinfo(skb)->nr_frags;
- for (i = 0; i < num_frags; i++) {
- skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
-
- if (length == 0) {
- /* don't need this page */
- skb_fill_page_desc(toskb, i, frag->page, 0, PAGE_SIZE);
- --skb_shinfo(skb)->nr_frags;
- } else {
- size = min(length, (unsigned) PAGE_SIZE);
-
- frag->size = size;
- skb->data_len += size;
- skb->truesize += size;
- skb->len += size;
- length -= size;
- }
- }
-}
void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
{
struct ipoib_dev_priv *priv = netdev_priv(dev);
- struct ipoib_cm_rx_buf *rx_ring;
+ struct ipoib_rx_buf *rx_ring;
unsigned int wr_id = wc->wr_id & ~(IPOIB_OP_CM | IPOIB_OP_RECV);
struct sk_buff *skb, *newskb;
struct ipoib_cm_rx *p;
@@ -582,7 +490,7 @@ void ipoib_cm_handle_rx_wc(struct net_de
frags = PAGE_ALIGN(wc->byte_len - min(wc->byte_len,
(unsigned)IPOIB_CM_HEAD_SIZE)) / PAGE_SIZE;
- newskb = ipoib_cm_alloc_rx_skb(dev, rx_ring, wr_id, frags, mapping);
+ newskb = ipoib_alloc_rx_skb(dev, wr_id, frags, IPOIB_CM_HEAD_SIZE, 12,
mapping);
if (unlikely(!newskb)) {
/*
* If we can't allocate a new RX buffer, dump
@@ -592,8 +500,9 @@ void ipoib_cm_handle_rx_wc(struct net_de
++dev->stats.rx_dropped;
goto repost;
}
+ rx_ring[wr_id].skb = newskb;
- ipoib_cm_dma_unmap_rx(priv, frags, rx_ring[wr_id].mapping);
+ ipoib_dma_unmap_rx(priv, frags, IPOIB_CM_HEAD_SIZE,
rx_ring[wr_id].mapping);
memcpy(rx_ring[wr_id].mapping, mapping, (frags + 1) * sizeof
*mapping);
ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
@@ -1508,9 +1417,10 @@ int ipoib_cm_dev_init(struct net_device
if (ipoib_cm_has_srq(dev)) {
for (i = 0; i < ipoib_recvq_size; ++i) {
- if (!ipoib_cm_alloc_rx_skb(dev, priv->cm.srq_ring, i,
- priv->cm.num_frags - 1,
- priv->cm.srq_ring[i].mapping)) {
+ priv->cm.srq_ring[i].skb = ipoib_alloc_rx_skb(dev, i,
priv->cm.num_frags - 1,
+ IPOIB_CM_HEAD_SIZE, 12,
+ priv->cm.srq_ring[i].mapping);
+ if (!priv->cm.srq_ring[i].skb) {
ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
ipoib_cm_dev_cleanup(dev);
return -ENOMEM;
diff -urpN
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib.h
ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib.h
--- ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib.h
2008-01-27 14:20:17.000000000 -0600
+++ ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib.h
2008-01-27 14:23:34.000000000 -0600
@@ -56,10 +56,9 @@
/* constants */
enum {
- IPOIB_PACKET_SIZE = 2048,
- IPOIB_BUF_SIZE = IPOIB_PACKET_SIZE + IB_GRH_BYTES,
-
IPOIB_ENCAP_LEN = 4,
+ IPOIB_MAX_IB_MTU = 4096, /* max ib device payload is 4096 */
+ IPOIB_UD_MAX_RX_SG = ALIGN(IPOIB_MAX_IB_MTU + IB_GRH_BYTES + 4,
PAGE_SIZE) / PAGE_SIZE, /* padding to align IP header */
IPOIB_CM_MTU = 0x10000 - 0x10, /* padding to align header
to 16 */
IPOIB_CM_BUF_SIZE = IPOIB_CM_MTU + IPOIB_ENCAP_LEN,
@@ -142,7 +141,7 @@ struct ipoib_mcast {
struct ipoib_rx_buf {
struct sk_buff *skb;
- u64 mapping;
+ u64 mapping[IPOIB_CM_RX_SG];
};
struct ipoib_tx_buf {
@@ -261,7 +260,7 @@ enum ipoib_cm_state {
struct ipoib_cm_rx {
struct ib_cm_id *id;
struct ib_qp *qp;
- struct ipoib_cm_rx_buf *rx_ring;
+ struct ipoib_rx_buf *rx_ring;
struct list_head list;
struct net_device *dev;
unsigned long jiffies;
@@ -285,14 +284,9 @@ struct ipoib_cm_tx {
struct ib_wc ibwc[IPOIB_NUM_WC];
};
-struct ipoib_cm_rx_buf {
- struct sk_buff *skb;
- u64 mapping[IPOIB_CM_RX_SG];
-};
-
struct ipoib_cm_dev_priv {
struct ib_srq *srq;
- struct ipoib_cm_rx_buf *srq_ring;
+ struct ipoib_rx_buf *srq_ring;
struct ib_cm_id *id;
struct list_head passive_ids; /* state: LIVE */
struct list_head rx_error_list; /* state: ERROR */
@@ -398,6 +392,9 @@ struct ipoib_dev_priv {
struct dentry *path_dentry;
#endif
struct ipoib_ethtool_st etool;
+ unsigned int max_ib_mtu;
+ struct ib_sge rx_sge[IPOIB_UD_MAX_RX_SG];
+ struct ib_recv_wr rx_wr;
};
struct ipoib_ah {
@@ -493,6 +490,14 @@ int ipoib_ib_dev_stop(struct net_device
int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int
port);
void ipoib_dev_cleanup(struct net_device *dev);
+void ipoib_dma_unmap_rx(struct ipoib_dev_priv *priv, int frags, int
head_size,
+ u64 *mapping);
+void skb_put_frags(struct sk_buff *skb, unsigned int hdr_space,
+ unsigned int length, struct sk_buff *toskb);
+struct sk_buff *ipoib_alloc_rx_skb(struct net_device *dev,
+ int id, int frags, int head_size,
+ int pad, u64 *mapping);
+
void ipoib_mcast_join_task(struct work_struct *work);
void ipoib_mcast_send(struct net_device *dev, void *mgid, struct
sk_buff *skb);
@@ -542,6 +547,11 @@ void ipoib_drain_cq(struct net_device *d
void ipoib_set_ethtool_ops(struct net_device *dev);
+#define IPOIB_UD_MTU(ib_mtu) (ib_mtu - IPOIB_ENCAP_LEN)
+#define IPOIB_UD_BUF_SIZE(ib_mtu) (ib_mtu + IB_GRH_BYTES + 4) /*
padding to align IP header */
+#define IPOIB_UD_HEAD_SIZE(ib_mtu) (IPOIB_UD_BUF_SIZE(ib_mtu)) %
PAGE_SIZE
+#define IPOIB_UD_RX_SG(ib_mtu) ALIGN(IPOIB_UD_BUF_SIZE(ib_mtu),
PAGE_SIZE) / PAGE_SIZE
+
#ifdef CONFIG_INFINIBAND_IPOIB_CM
#define IPOIB_FLAGS_RC 0x80
diff -urpN
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
---
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
2008-01-27 14:20:17.000000000 -0600
+++
ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
2008-01-27 14:26:05.000000000 -0600
@@ -89,63 +89,118 @@ void ipoib_free_ah(struct kref *kref)
spin_unlock_irqrestore(&priv->lock, flags);
}
-static int ipoib_ib_post_receive(struct net_device *dev, int id)
+/* Adjust length of skb with fragments to match received data */
+void skb_put_frags(struct sk_buff *skb, unsigned int hdr_space,
+ unsigned int length, struct sk_buff *toskb)
{
- struct ipoib_dev_priv *priv = netdev_priv(dev);
- struct ib_sge list;
- struct ib_recv_wr param;
- struct ib_recv_wr *bad_wr;
- int ret;
+ int i, num_frags;
+ unsigned int size;
- list.addr = priv->rx_ring[id].mapping;
- list.length = IPOIB_BUF_SIZE;
- list.lkey = priv->mr->lkey;
-
- param.next = NULL;
- param.wr_id = id | IPOIB_OP_RECV;
- param.sg_list = &list;
- param.num_sge = 1;
+ /* put header into skb */
+ size = min(length, hdr_space);
+ skb->tail += size;
+ skb->len += size;
+ length -= size;
- ret = ib_post_recv(priv->qp, ¶m, &bad_wr);
- if (unlikely(ret)) {
- ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret);
- ib_dma_unmap_single(priv->ca, priv->rx_ring[id].mapping,
- IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
- dev_kfree_skb_any(priv->rx_ring[id].skb);
- priv->rx_ring[id].skb = NULL;
+ num_frags = skb_shinfo(skb)->nr_frags;
+ for (i = 0; i < num_frags; i++) {
+ skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+
+ if (length == 0) {
+ /* don't need this page */
+ skb_fill_page_desc(toskb, i, frag->page, 0, PAGE_SIZE);
+ --skb_shinfo(skb)->nr_frags;
+ } else {
+ size = min(length, (unsigned) PAGE_SIZE);
+
+ frag->size = size;
+ skb->data_len += size;
+ skb->truesize += size;
+ skb->len += size;
+ length -= size;
+ }
}
+}
- return ret;
+void ipoib_dma_unmap_rx(struct ipoib_dev_priv *priv, int frags, int
head_size, u64 *mapping)
+{
+ int i;
+ ib_dma_unmap_single(priv->ca, mapping[0], head_size,
+ DMA_FROM_DEVICE);
+ for (i = 0; i < frags; i++)
+ ib_dma_unmap_single(priv->ca, mapping[i+1], PAGE_SIZE,
+ DMA_FROM_DEVICE);
}
-static int ipoib_alloc_rx_skb(struct net_device *dev, int id)
+struct sk_buff *ipoib_alloc_rx_skb(struct net_device *dev, int id, int
frags,
+ int head_size, int pad, u64 *mapping)
{
struct ipoib_dev_priv *priv = netdev_priv(dev);
struct sk_buff *skb;
- u64 addr;
+ int i;
- skb = dev_alloc_skb(IPOIB_BUF_SIZE + 4);
- if (!skb)
- return -ENOMEM;
+ skb = dev_alloc_skb(head_size + pad);
+ if (unlikely(!skb))
+ return NULL;
/*
- * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte
- * header. So we need 4 more bytes to get to 48 and align the
+ * IPoIB adds a 4 byte header. So we need 12 more bytes to align the
* IP header to a multiple of 16.
*/
- skb_reserve(skb, 4);
+ skb_reserve(skb, pad);
- addr = ib_dma_map_single(priv->ca, skb->data, IPOIB_BUF_SIZE,
- DMA_FROM_DEVICE);
- if (unlikely(ib_dma_mapping_error(priv->ca, addr))) {
+ mapping[0] = ib_dma_map_single(priv->ca, skb->data, head_size,
+ DMA_FROM_DEVICE);
+ if (unlikely(ib_dma_mapping_error(priv->ca, mapping[0]))) {
dev_kfree_skb_any(skb);
- return -EIO;
+ return NULL;
}
- priv->rx_ring[id].skb = skb;
- priv->rx_ring[id].mapping = addr;
+ for (i = 0; i < frags; i++) {
+ struct page *page = alloc_page(GFP_ATOMIC);
- return 0;
+ if (!page)
+ goto partial_error;
+ skb_fill_page_desc(skb, i, page, 0, PAGE_SIZE);
+
+ mapping[i + 1] = ib_dma_map_page(priv->ca,
skb_shinfo(skb)->frags[i].page,
+ 0, PAGE_SIZE, DMA_FROM_DEVICE);
+ if (unlikely(ib_dma_mapping_error(priv->ca, mapping[i + 1])))
+ goto partial_error;
+ }
+
+ return skb;
+
+partial_error:
+
+ ib_dma_unmap_single(priv->ca, mapping[0], head_size, DMA_FROM_DEVICE);
+
+ for (; i > 0; --i)
+ ib_dma_unmap_single(priv->ca, mapping[i], PAGE_SIZE,
DMA_FROM_DEVICE);
+
+ dev_kfree_skb_any(skb);
+ return NULL;
+}
+
+static int ipoib_ib_post_receive(struct net_device *dev, int id)
+{
+ struct ipoib_dev_priv *priv = netdev_priv(dev);
+ struct ib_recv_wr *bad_wr;
+ int ret, i;
+
+ priv->rx_wr.wr_id = id | IPOIB_OP_RECV;
+ for (i = 0; i < IPOIB_UD_RX_SG(priv->max_ib_mtu); ++i)
+ priv->rx_sge[i].addr = priv->rx_ring[id].mapping[i];
+ ret = ib_post_recv(priv->qp, &priv->rx_wr, &bad_wr);
+ if (unlikely(ret)) {
+ ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret);
+ ipoib_dma_unmap_rx(priv, IPOIB_UD_RX_SG(priv->max_ib_mtu) - 1,
+ IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu),
priv->rx_ring[id].mapping);
+ dev_kfree_skb_any(priv->rx_ring[id].skb);
+ priv->rx_ring[id].skb = NULL;
+ }
+
+ return ret;
}
static int ipoib_ib_post_receives(struct net_device *dev)
@@ -153,13 +208,24 @@ static int ipoib_ib_post_receives(struct
struct ipoib_dev_priv *priv = netdev_priv(dev);
int i;
+ for (i = 0; i < IPOIB_UD_RX_SG(priv->max_ib_mtu); ++i)
+ priv->rx_sge[i].lkey = priv->mr->lkey;
+ priv->rx_sge[0].length = IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu);
+ for (i = 0; i < IPOIB_UD_RX_SG(priv->max_ib_mtu) - 1; ++i)
+ priv->rx_sge[i+1].length = PAGE_SIZE;
+ priv->rx_wr.num_sge = IPOIB_UD_RX_SG(priv->max_ib_mtu);
+ priv->rx_wr.next = NULL;
+ priv->rx_wr.sg_list = priv->rx_sge;
+
for (i = 0; i < ipoib_recvq_size; ++i) {
- if (ipoib_alloc_rx_skb(dev, i)) {
- ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
+ priv->rx_ring[i].skb = ipoib_alloc_rx_skb(dev, i,
IPOIB_UD_RX_SG(priv->max_ib_mtu) - 1,
+ IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu), 4,
+ priv->rx_ring[i].mapping);
+ if (!priv->rx_ring[i].skb)
return -ENOMEM;
- }
if (ipoib_ib_post_receive(dev, i)) {
ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i);
+ ipoib_dev_cleanup(dev);
return -EIO;
}
}
@@ -171,9 +237,10 @@ static void ipoib_ib_handle_rx_wc(struct
{
struct ipoib_dev_priv *priv = netdev_priv(dev);
unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV;
- struct sk_buff *skb;
+ struct sk_buff *skb, *newskb;
+ u64 mapping[IPOIB_UD_RX_SG(priv->max_ib_mtu)];
struct ipoib_header *header;
- u64 addr;
+ int frags;
ipoib_dbg_data(priv, "recv completion: id %d, status: %d\n",
wr_id, wc->status);
@@ -185,15 +252,15 @@ static void ipoib_ib_handle_rx_wc(struct
}
skb = priv->rx_ring[wr_id].skb;
- addr = priv->rx_ring[wr_id].mapping;
if (unlikely(wc->status != IB_WC_SUCCESS)) {
if (wc->status != IB_WC_WR_FLUSH_ERR)
ipoib_warn(priv, "failed recv event "
"(status=%d, wrid=%d vend_err %x)\n",
wc->status, wr_id, wc->vendor_err);
- ib_dma_unmap_single(priv->ca, addr,
- IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
+ ipoib_dma_unmap_rx(priv, IPOIB_UD_RX_SG(priv->max_ib_mtu) - 1,
+ IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu),
+ priv->rx_ring[wr_id].mapping);
dev_kfree_skb_any(skb);
priv->rx_ring[wr_id].skb = NULL;
return;
@@ -206,21 +273,28 @@ static void ipoib_ib_handle_rx_wc(struct
if (wc->slid == priv->local_lid && wc->src_qp == priv->qp->qp_num)
goto repost;
+ frags = PAGE_ALIGN(wc->byte_len - min(wc->byte_len,
+ (unsigned)IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu))) /
PAGE_SIZE;
+ newskb = ipoib_alloc_rx_skb(dev, wr_id, frags,
+ IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu), 4, mapping);
/*
* If we can't allocate a new RX buffer, dump
* this packet and reuse the old buffer.
*/
- if (unlikely(ipoib_alloc_rx_skb(dev, wr_id))) {
- ++dev->stats.rx_dropped;
- goto repost;
- }
+ if (unlikely(!newskb)) {
+ ipoib_dbg(priv, "failed to allocate receive buffer %d\n", wr_id);
+ ++dev->stats.rx_dropped;
+ goto repost;
+ }
+ priv->rx_ring[wr_id].skb = newskb;
ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
wc->byte_len, wc->slid);
- ib_dma_unmap_single(priv->ca, addr, IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
+ ipoib_dma_unmap_rx(priv, frags, IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu),
priv->rx_ring[wr_id].mapping);
+ memcpy(priv->rx_ring[wr_id].mapping, mapping, (frags + 1) * sizeof
*mapping);
- skb_put(skb, wc->byte_len);
+ skb_put_frags(skb, IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu), wc->byte_len,
newskb);
skb_pull(skb, IB_GRH_BYTES);
header = (struct ipoib_header *)skb->data;
@@ -687,10 +761,10 @@ int ipoib_ib_dev_stop(struct net_device
rx_req = &priv->rx_ring[i];
if (!rx_req->skb)
continue;
- ib_dma_unmap_single(priv->ca,
- rx_req->mapping,
- IPOIB_BUF_SIZE,
- DMA_FROM_DEVICE);
+ ipoib_dma_unmap_rx(priv,
+ IPOIB_UD_RX_SG(priv->max_ib_mtu) - 1,
+ IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu),
+ priv->rx_ring[i].mapping);
dev_kfree_skb_any(rx_req->skb);
rx_req->skb = NULL;
}
diff -urpN
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_main.c
ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_main.c
---
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_main.c
2008-01-27 14:20:17.000000000 -0600
+++
ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_main.c
2008-01-27 14:23:34.000000000 -0600
@@ -196,7 +196,7 @@ static int ipoib_change_mtu(struct net_d
return 0;
}
- if (new_mtu > IPOIB_PACKET_SIZE - IPOIB_ENCAP_LEN)
+ if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu))
return -EINVAL;
priv->admin_mtu = new_mtu;
@@ -1024,10 +1024,6 @@ static void ipoib_setup(struct net_devic
set_bit(IPOIB_FLAG_HW_CSUM, &priv->flags);
}
- /* MTU will be reset when mcast join happens */
- dev->mtu = IPOIB_PACKET_SIZE - IPOIB_ENCAP_LEN;
- priv->mcast_mtu = priv->admin_mtu = dev->mtu;
-
memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
netif_carrier_off(dev);
@@ -1182,6 +1178,7 @@ static struct net_device *ipoib_add_port
struct ib_device *hca, u8 port)
{
struct ipoib_dev_priv *priv;
+ struct ib_port_attr attr;
int result = -ENOMEM;
priv = ipoib_intf_alloc(format);
@@ -1192,6 +1189,18 @@ static struct net_device *ipoib_add_port
priv->dev->features |= NETIF_F_HIGHDMA;
+ if (!ib_query_port(hca, port, &attr))
+ priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu);
+ else {
+ printk(KERN_WARNING "%s: ib_query_port %d failed\n",
+ hca->name, port);
+ goto device_init_failed;
+ }
+
+ /* MTU will be reset when mcast join happens */
+ priv->dev->mtu = IPOIB_UD_MTU(priv->max_ib_mtu);
+ priv->mcast_mtu = priv->admin_mtu = priv->dev->mtu;
+
result = ib_query_pkey(hca, port, 0, &priv->pkey);
if (result) {
printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
diff -urpN
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
---
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c 2008-01-14 04:00:04.000000000 -0600
+++
ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c 2008-01-27 14:23:34.000000000 -0600
@@ -567,9 +567,7 @@ void ipoib_mcast_join_task(struct work_s
return;
}
- priv->mcast_mtu = ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu) -
- IPOIB_ENCAP_LEN;
-
+ 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);
diff -urpN
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
---
ofa_1_3_kernel-20080114-0200_a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c 2008-01-27 14:20:17.000000000 -0600
+++
ofa_1_3_kernel-20080114-0200_b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c 2008-01-27 14:23:34.000000000 -0600
@@ -150,7 +150,7 @@ int ipoib_transport_dev_init(struct net_
.max_send_wr = ipoib_sendq_size,
.max_recv_wr = ipoib_recvq_size,
.max_send_sge = dev->features & NETIF_F_SG ? MAX_SKB_FRAGS + 1 : 1,
- .max_recv_sge = 1
+ .max_recv_sge = IPOIB_UD_RX_SG(priv->max_ib_mtu)
},
.sq_sig_type = IB_SIGNAL_ALL_WR,
.qp_type = IB_QPT_UD,
@@ -212,6 +212,16 @@ int ipoib_transport_dev_init(struct net_
priv->tx_wr.sg_list = priv->tx_sge;
priv->tx_wr.send_flags = IB_SEND_SIGNALED;
+ priv->rx_sge[0].length = IPOIB_UD_HEAD_SIZE(priv->max_ib_mtu);
+ for (i = 0; i < IPOIB_UD_RX_SG(priv->max_ib_mtu) - 1; ++i) {
+ priv->rx_sge[i].lkey = priv->mr->lkey;
+ priv->rx_sge[i + 1].length = PAGE_SIZE;
+ }
+ priv->rx_sge[i + 1].lkey = priv->mr->lkey;
+ priv->rx_wr.num_sge = IPOIB_UD_RX_SG(priv->max_ib_mtu);
+ priv->rx_wr.next = NULL;
+ priv->rx_wr.sg_list = priv->rx_sge;
+
return 0;
out_free_cq:
More information about the general
mailing list