summaryrefslogtreecommitdiff
path: root/drivers/infiniband/core/mad.c
diff options
context:
space:
mode:
authorJack Morgenstein <jackm@mellanox.co.il>2006-03-03 21:54:13 -0800
committerRoland Dreier <rolandd@cisco.com>2006-03-20 10:08:23 -0800
commitf36e1793e25513380cae5958a9164d4cc4458ad0 (patch)
treeaa31d34ee07971645af6f21068709166420caee2 /drivers/infiniband/core/mad.c
parent6ecb0c849625e830ab96495d473bb704812c30e1 (diff)
IB/umad: Add support for large RMPP transfers
Add support for sending and receiving large RMPP transfers. The old code supports transfers only as large as a single contiguous kernel memory allocation. This patch uses linked list of memory buffers when sending and receiving data to avoid needing contiguous pages for larger transfers. Receive side: copy the arriving MADs in chunks instead of coalescing to one large buffer in kernel space. Send side: split a multipacket MAD buffer to a list of segments, (multipacket_list) and send these using a gather list of size 2. Also, save pointer to last sent segment, and retrieve requested segments by walking list starting at last sent segment. Finally, save pointer to last-acked segment. When retrying, retrieve segments for resending relative to this pointer. When updating last ack, start at this pointer. Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il> Signed-off-by: Sean Hefty <sean.hefty@intel.com> Signed-off-by: Roland Dreier <rolandd@cisco.com>
Diffstat (limited to 'drivers/infiniband/core/mad.c')
-rw-r--r--drivers/infiniband/core/mad.c166
1 files changed, 135 insertions, 31 deletions
diff --git a/drivers/infiniband/core/mad.c b/drivers/infiniband/core/mad.c
index 445ad0dda213..16549add8e8f 100644
--- a/drivers/infiniband/core/mad.c
+++ b/drivers/infiniband/core/mad.c
@@ -31,7 +31,7 @@
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
- * $Id: mad.c 2817 2005-07-07 11:29:26Z halr $
+ * $Id: mad.c 5596 2006-03-03 01:00:07Z sean.hefty $
*/
#include <linux/dma-mapping.h>
@@ -765,18 +765,67 @@ out:
return ret;
}
-static int get_buf_length(int hdr_len, int data_len)
+static int get_pad_size(int hdr_len, int data_len)
{
int seg_size, pad;
seg_size = sizeof(struct ib_mad) - hdr_len;
if (data_len && seg_size) {
pad = seg_size - data_len % seg_size;
- if (pad == seg_size)
- pad = 0;
+ return pad == seg_size ? 0 : pad;
} else
- pad = seg_size;
- return hdr_len + data_len + pad;
+ return seg_size;
+}
+
+static void free_send_rmpp_list(struct ib_mad_send_wr_private *mad_send_wr)
+{
+ struct ib_rmpp_segment *s, *t;
+
+ list_for_each_entry_safe(s, t, &mad_send_wr->rmpp_list, list) {
+ list_del(&s->list);
+ kfree(s);
+ }
+}
+
+static int alloc_send_rmpp_list(struct ib_mad_send_wr_private *send_wr,
+ gfp_t gfp_mask)
+{
+ struct ib_mad_send_buf *send_buf = &send_wr->send_buf;
+ struct ib_rmpp_mad *rmpp_mad = send_buf->mad;
+ struct ib_rmpp_segment *seg = NULL;
+ int left, seg_size, pad;
+
+ send_buf->seg_size = sizeof (struct ib_mad) - send_buf->hdr_len;
+ seg_size = send_buf->seg_size;
+ pad = send_wr->pad;
+
+ /* Allocate data segments. */
+ for (left = send_buf->data_len + pad; left > 0; left -= seg_size) {
+ seg = kmalloc(sizeof (*seg) + seg_size, gfp_mask);
+ if (!seg) {
+ printk(KERN_ERR "alloc_send_rmpp_segs: RMPP mem "
+ "alloc failed for len %zd, gfp %#x\n",
+ sizeof (*seg) + seg_size, gfp_mask);
+ free_send_rmpp_list(send_wr);
+ return -ENOMEM;
+ }
+ seg->num = ++send_buf->seg_count;
+ list_add_tail(&seg->list, &send_wr->rmpp_list);
+ }
+
+ /* Zero any padding */
+ if (pad)
+ memset(seg->data + seg_size - pad, 0, pad);
+
+ rmpp_mad->rmpp_hdr.rmpp_version = send_wr->mad_agent_priv->
+ agent.rmpp_version;
+ rmpp_mad->rmpp_hdr.rmpp_type = IB_MGMT_RMPP_TYPE_DATA;
+ ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE);
+
+ send_wr->cur_seg = container_of(send_wr->rmpp_list.next,
+ struct ib_rmpp_segment, list);
+ send_wr->last_ack_seg = send_wr->cur_seg;
+ return 0;
}
struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
@@ -787,32 +836,40 @@ struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
{
struct ib_mad_agent_private *mad_agent_priv;
struct ib_mad_send_wr_private *mad_send_wr;
- int buf_size;
+ int pad, message_size, ret, size;
void *buf;
mad_agent_priv = container_of(mad_agent, struct ib_mad_agent_private,
agent);
- buf_size = get_buf_length(hdr_len, data_len);
+ pad = get_pad_size(hdr_len, data_len);
+ message_size = hdr_len + data_len + pad;
if ((!mad_agent->rmpp_version &&
- (rmpp_active || buf_size > sizeof(struct ib_mad))) ||
- (!rmpp_active && buf_size > sizeof(struct ib_mad)))
+ (rmpp_active || message_size > sizeof(struct ib_mad))) ||
+ (!rmpp_active && message_size > sizeof(struct ib_mad)))
return ERR_PTR(-EINVAL);
- buf = kzalloc(sizeof *mad_send_wr + buf_size, gfp_mask);
+ size = rmpp_active ? hdr_len : sizeof(struct ib_mad);
+ buf = kzalloc(sizeof *mad_send_wr + size, gfp_mask);
if (!buf)
return ERR_PTR(-ENOMEM);
- mad_send_wr = buf + buf_size;
+ mad_send_wr = buf + size;
+ INIT_LIST_HEAD(&mad_send_wr->rmpp_list);
mad_send_wr->send_buf.mad = buf;
+ mad_send_wr->send_buf.hdr_len = hdr_len;
+ mad_send_wr->send_buf.data_len = data_len;
+ mad_send_wr->pad = pad;
mad_send_wr->mad_agent_priv = mad_agent_priv;
- mad_send_wr->sg_list[0].length = buf_size;
+ mad_send_wr->sg_list[0].length = hdr_len;
mad_send_wr->sg_list[0].lkey = mad_agent->mr->lkey;
+ mad_send_wr->sg_list[1].length = sizeof(struct ib_mad) - hdr_len;
+ mad_send_wr->sg_list[1].lkey = mad_agent->mr->lkey;
mad_send_wr->send_wr.wr_id = (unsigned long) mad_send_wr;
mad_send_wr->send_wr.sg_list = mad_send_wr->sg_list;
- mad_send_wr->send_wr.num_sge = 1;
+ mad_send_wr->send_wr.num_sge = 2;
mad_send_wr->send_wr.opcode = IB_WR_SEND;
mad_send_wr->send_wr.send_flags = IB_SEND_SIGNALED;
mad_send_wr->send_wr.wr.ud.remote_qpn = remote_qpn;
@@ -820,13 +877,11 @@ struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
mad_send_wr->send_wr.wr.ud.pkey_index = pkey_index;
if (rmpp_active) {
- struct ib_rmpp_mad *rmpp_mad = mad_send_wr->send_buf.mad;
- rmpp_mad->rmpp_hdr.paylen_newwin = cpu_to_be32(hdr_len -
- IB_MGMT_RMPP_HDR + data_len);
- rmpp_mad->rmpp_hdr.rmpp_version = mad_agent->rmpp_version;
- rmpp_mad->rmpp_hdr.rmpp_type = IB_MGMT_RMPP_TYPE_DATA;
- ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr,
- IB_MGMT_RMPP_FLAG_ACTIVE);
+ ret = alloc_send_rmpp_list(mad_send_wr, gfp_mask);
+ if (ret) {
+ kfree(buf);
+ return ERR_PTR(ret);
+ }
}
mad_send_wr->send_buf.mad_agent = mad_agent;
@@ -835,14 +890,50 @@ struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
}
EXPORT_SYMBOL(ib_create_send_mad);
+void *ib_get_rmpp_segment(struct ib_mad_send_buf *send_buf, int seg_num)
+{
+ struct ib_mad_send_wr_private *mad_send_wr;
+ struct list_head *list;
+
+ mad_send_wr = container_of(send_buf, struct ib_mad_send_wr_private,
+ send_buf);
+ list = &mad_send_wr->cur_seg->list;
+
+ if (mad_send_wr->cur_seg->num < seg_num) {
+ list_for_each_entry(mad_send_wr->cur_seg, list, list)
+ if (mad_send_wr->cur_seg->num == seg_num)
+ break;
+ } else if (mad_send_wr->cur_seg->num > seg_num) {
+ list_for_each_entry_reverse(mad_send_wr->cur_seg, list, list)
+ if (mad_send_wr->cur_seg->num == seg_num)
+ break;
+ }
+ return mad_send_wr->cur_seg->data;
+}
+EXPORT_SYMBOL(ib_get_rmpp_segment);
+
+static inline void *ib_get_payload(struct ib_mad_send_wr_private *mad_send_wr)
+{
+ if (mad_send_wr->send_buf.seg_count)
+ return ib_get_rmpp_segment(&mad_send_wr->send_buf,
+ mad_send_wr->seg_num);
+ else
+ return mad_send_wr->send_buf.mad +
+ mad_send_wr->send_buf.hdr_len;
+}
+
void ib_free_send_mad(struct ib_mad_send_buf *send_buf)
{
struct ib_mad_agent_private *mad_agent_priv;
+ struct ib_mad_send_wr_private *mad_send_wr;
mad_agent_priv = container_of(send_buf->mad_agent,
struct ib_mad_agent_private, agent);
- kfree(send_buf->mad);
+ mad_send_wr = container_of(send_buf, struct ib_mad_send_wr_private,
+ send_buf);
+ free_send_rmpp_list(mad_send_wr);
+ kfree(send_buf->mad);
if (atomic_dec_and_test(&mad_agent_priv->refcount))
wake_up(&mad_agent_priv->wait);
}
@@ -865,10 +956,17 @@ int ib_send_mad(struct ib_mad_send_wr_private *mad_send_wr)
mad_agent = mad_send_wr->send_buf.mad_agent;
sge = mad_send_wr->sg_list;
- sge->addr = dma_map_single(mad_agent->device->dma_device,
- mad_send_wr->send_buf.mad, sge->length,
- DMA_TO_DEVICE);
- pci_unmap_addr_set(mad_send_wr, mapping, sge->addr);
+ sge[0].addr = dma_map_single(mad_agent->device->dma_device,
+ mad_send_wr->send_buf.mad,
+ sge[0].length,
+ DMA_TO_DEVICE);
+ pci_unmap_addr_set(mad_send_wr, header_mapping, sge[0].addr);
+
+ sge[1].addr = dma_map_single(mad_agent->device->dma_device,
+ ib_get_payload(mad_send_wr),
+ sge[1].length,
+ DMA_TO_DEVICE);
+ pci_unmap_addr_set(mad_send_wr, payload_mapping, sge[1].addr);
spin_lock_irqsave(&qp_info->send_queue.lock, flags);
if (qp_info->send_queue.count < qp_info->send_queue.max_active) {
@@ -885,11 +983,14 @@ int ib_send_mad(struct ib_mad_send_wr_private *mad_send_wr)
list_add_tail(&mad_send_wr->mad_list.list, list);
}
spin_unlock_irqrestore(&qp_info->send_queue.lock, flags);
- if (ret)
+ if (ret) {
dma_unmap_single(mad_agent->device->dma_device,
- pci_unmap_addr(mad_send_wr, mapping),
- sge->length, DMA_TO_DEVICE);
-
+ pci_unmap_addr(mad_send_wr, header_mapping),
+ sge[0].length, DMA_TO_DEVICE);
+ dma_unmap_single(mad_agent->device->dma_device,
+ pci_unmap_addr(mad_send_wr, payload_mapping),
+ sge[1].length, DMA_TO_DEVICE);
+ }
return ret;
}
@@ -1860,8 +1961,11 @@ static void ib_mad_send_done_handler(struct ib_mad_port_private *port_priv,
retry:
dma_unmap_single(mad_send_wr->send_buf.mad_agent->device->dma_device,
- pci_unmap_addr(mad_send_wr, mapping),
+ pci_unmap_addr(mad_send_wr, header_mapping),
mad_send_wr->sg_list[0].length, DMA_TO_DEVICE);
+ dma_unmap_single(mad_send_wr->send_buf.mad_agent->device->dma_device,
+ pci_unmap_addr(mad_send_wr, payload_mapping),
+ mad_send_wr->sg_list[1].length, DMA_TO_DEVICE);
queued_send_wr = NULL;
spin_lock_irqsave(&send_queue->lock, flags);
list_del(&mad_list->list);