summaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/mthca/mthca_qp.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/infiniband/hw/mthca/mthca_qp.c')
-rw-r--r--drivers/infiniband/hw/mthca/mthca_qp.c344
1 files changed, 246 insertions, 98 deletions
diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index ca73bab11a02..f7126b14d5ae 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -1,5 +1,6 @@
/*
* Copyright (c) 2004 Topspin Communications. All rights reserved.
+ * Copyright (c) 2005 Cisco Systems. All rights reserved.
*
* This software is available to you under a choice of one of two
* licenses. You may choose to be licensed under the terms of the GNU
@@ -46,7 +47,9 @@ enum {
MTHCA_MAX_DIRECT_QP_SIZE = 4 * PAGE_SIZE,
MTHCA_ACK_REQ_FREQ = 10,
MTHCA_FLIGHT_LIMIT = 9,
- MTHCA_UD_HEADER_SIZE = 72 /* largest UD header possible */
+ MTHCA_UD_HEADER_SIZE = 72, /* largest UD header possible */
+ MTHCA_INLINE_HEADER_SIZE = 4, /* data segment overhead for inline */
+ MTHCA_INLINE_CHUNK_SIZE = 16 /* inline data segment chunk */
};
enum {
@@ -357,6 +360,9 @@ static const struct {
[UD] = (IB_QP_PKEY_INDEX |
IB_QP_PORT |
IB_QP_QKEY),
+ [UC] = (IB_QP_PKEY_INDEX |
+ IB_QP_PORT |
+ IB_QP_ACCESS_FLAGS),
[RC] = (IB_QP_PKEY_INDEX |
IB_QP_PORT |
IB_QP_ACCESS_FLAGS),
@@ -378,6 +384,9 @@ static const struct {
[UD] = (IB_QP_PKEY_INDEX |
IB_QP_PORT |
IB_QP_QKEY),
+ [UC] = (IB_QP_PKEY_INDEX |
+ IB_QP_PORT |
+ IB_QP_ACCESS_FLAGS),
[RC] = (IB_QP_PKEY_INDEX |
IB_QP_PORT |
IB_QP_ACCESS_FLAGS),
@@ -388,6 +397,11 @@ static const struct {
[IB_QPS_RTR] = {
.trans = MTHCA_TRANS_INIT2RTR,
.req_param = {
+ [UC] = (IB_QP_AV |
+ IB_QP_PATH_MTU |
+ IB_QP_DEST_QPN |
+ IB_QP_RQ_PSN |
+ IB_QP_MAX_DEST_RD_ATOMIC),
[RC] = (IB_QP_AV |
IB_QP_PATH_MTU |
IB_QP_DEST_QPN |
@@ -398,6 +412,9 @@ static const struct {
.opt_param = {
[UD] = (IB_QP_PKEY_INDEX |
IB_QP_QKEY),
+ [UC] = (IB_QP_ALT_PATH |
+ IB_QP_ACCESS_FLAGS |
+ IB_QP_PKEY_INDEX),
[RC] = (IB_QP_ALT_PATH |
IB_QP_ACCESS_FLAGS |
IB_QP_PKEY_INDEX),
@@ -413,6 +430,8 @@ static const struct {
.trans = MTHCA_TRANS_RTR2RTS,
.req_param = {
[UD] = IB_QP_SQ_PSN,
+ [UC] = (IB_QP_SQ_PSN |
+ IB_QP_MAX_QP_RD_ATOMIC),
[RC] = (IB_QP_TIMEOUT |
IB_QP_RETRY_CNT |
IB_QP_RNR_RETRY |
@@ -423,6 +442,11 @@ static const struct {
.opt_param = {
[UD] = (IB_QP_CUR_STATE |
IB_QP_QKEY),
+ [UC] = (IB_QP_CUR_STATE |
+ IB_QP_ALT_PATH |
+ IB_QP_ACCESS_FLAGS |
+ IB_QP_PKEY_INDEX |
+ IB_QP_PATH_MIG_STATE),
[RC] = (IB_QP_CUR_STATE |
IB_QP_ALT_PATH |
IB_QP_ACCESS_FLAGS |
@@ -442,6 +466,9 @@ static const struct {
.opt_param = {
[UD] = (IB_QP_CUR_STATE |
IB_QP_QKEY),
+ [UC] = (IB_QP_ACCESS_FLAGS |
+ IB_QP_ALT_PATH |
+ IB_QP_PATH_MIG_STATE),
[RC] = (IB_QP_ACCESS_FLAGS |
IB_QP_ALT_PATH |
IB_QP_PATH_MIG_STATE |
@@ -462,6 +489,10 @@ static const struct {
.opt_param = {
[UD] = (IB_QP_CUR_STATE |
IB_QP_QKEY),
+ [UC] = (IB_QP_CUR_STATE |
+ IB_QP_ALT_PATH |
+ IB_QP_ACCESS_FLAGS |
+ IB_QP_PATH_MIG_STATE),
[RC] = (IB_QP_CUR_STATE |
IB_QP_ALT_PATH |
IB_QP_ACCESS_FLAGS |
@@ -476,6 +507,14 @@ static const struct {
.opt_param = {
[UD] = (IB_QP_PKEY_INDEX |
IB_QP_QKEY),
+ [UC] = (IB_QP_AV |
+ IB_QP_MAX_QP_RD_ATOMIC |
+ IB_QP_MAX_DEST_RD_ATOMIC |
+ IB_QP_CUR_STATE |
+ IB_QP_ALT_PATH |
+ IB_QP_ACCESS_FLAGS |
+ IB_QP_PKEY_INDEX |
+ IB_QP_PATH_MIG_STATE),
[RC] = (IB_QP_AV |
IB_QP_TIMEOUT |
IB_QP_RETRY_CNT |
@@ -501,6 +540,7 @@ static const struct {
.opt_param = {
[UD] = (IB_QP_CUR_STATE |
IB_QP_QKEY),
+ [UC] = (IB_QP_CUR_STATE),
[RC] = (IB_QP_CUR_STATE |
IB_QP_MIN_RNR_TIMER),
[MLX] = (IB_QP_CUR_STATE |
@@ -552,7 +592,7 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
struct mthca_dev *dev = to_mdev(ibqp->device);
struct mthca_qp *qp = to_mqp(ibqp);
enum ib_qp_state cur_state, new_state;
- void *mailbox = NULL;
+ struct mthca_mailbox *mailbox;
struct mthca_qp_param *qp_param;
struct mthca_qp_context *qp_context;
u32 req_param, opt_param;
@@ -609,10 +649,10 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
return -EINVAL;
}
- mailbox = kmalloc(sizeof (*qp_param) + MTHCA_CMD_MAILBOX_EXTRA, GFP_KERNEL);
- if (!mailbox)
- return -ENOMEM;
- qp_param = MAILBOX_ALIGN(mailbox);
+ mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
+ if (IS_ERR(mailbox))
+ return PTR_ERR(mailbox);
+ qp_param = mailbox->buf;
qp_context = &qp_param->context;
memset(qp_param, 0, sizeof *qp_param);
@@ -652,7 +692,11 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
/* leave arbel_sched_queue as 0 */
- qp_context->usr_page = cpu_to_be32(dev->driver_uar.index);
+ if (qp->ibqp.uobject)
+ qp_context->usr_page =
+ cpu_to_be32(to_mucontext(qp->ibqp.uobject->context)->uar.index);
+ else
+ qp_context->usr_page = cpu_to_be32(dev->driver_uar.index);
qp_context->local_qpn = cpu_to_be32(qp->qpn);
if (attr_mask & IB_QP_DEST_QPN) {
qp_context->remote_qpn = cpu_to_be32(attr->dest_qp_num);
@@ -683,7 +727,7 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
if (attr_mask & IB_QP_AV) {
qp_context->pri_path.g_mylmc = attr->ah_attr.src_path_bits & 0x7f;
qp_context->pri_path.rlid = cpu_to_be16(attr->ah_attr.dlid);
- qp_context->pri_path.static_rate = (!!attr->ah_attr.static_rate) << 3;
+ qp_context->pri_path.static_rate = !!attr->ah_attr.static_rate;
if (attr->ah_attr.ah_flags & IB_AH_GRH) {
qp_context->pri_path.g_mylmc |= 1 << 7;
qp_context->pri_path.mgid_index = attr->ah_attr.grh.sgid_index;
@@ -724,9 +768,9 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
qp_param->opt_param_mask |= cpu_to_be32(MTHCA_QP_OPTPAR_RETRY_COUNT);
}
- if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
- qp_context->params1 |= cpu_to_be32(min(attr->max_dest_rd_atomic ?
- ffs(attr->max_dest_rd_atomic) - 1 : 0,
+ if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
+ qp_context->params1 |= cpu_to_be32(min(attr->max_rd_atomic ?
+ ffs(attr->max_rd_atomic) - 1 : 0,
7) << 21);
qp_param->opt_param_mask |= cpu_to_be32(MTHCA_QP_OPTPAR_SRA_MAX);
}
@@ -764,10 +808,10 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
qp->atomic_rd_en = attr->qp_access_flags;
}
- if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
+ if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
u8 rra_max;
- if (qp->resp_depth && !attr->max_rd_atomic) {
+ if (qp->resp_depth && !attr->max_dest_rd_atomic) {
/*
* Lowering our responder resources to zero.
* Turn off RDMA/atomics as responder.
@@ -778,7 +822,7 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
MTHCA_QP_OPTPAR_RAE);
}
- if (!qp->resp_depth && attr->max_rd_atomic) {
+ if (!qp->resp_depth && attr->max_dest_rd_atomic) {
/*
* Increasing our responder resources from
* zero. Turn on RDMA/atomics as appropriate.
@@ -799,7 +843,7 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
}
for (rra_max = 0;
- 1 << rra_max < attr->max_rd_atomic &&
+ 1 << rra_max < attr->max_dest_rd_atomic &&
rra_max < dev->qp_table.rdb_shift;
++rra_max)
; /* nothing */
@@ -807,7 +851,7 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
qp_context->params2 |= cpu_to_be32(rra_max << 21);
qp_param->opt_param_mask |= cpu_to_be32(MTHCA_QP_OPTPAR_RRA_MAX);
- qp->resp_depth = attr->max_rd_atomic;
+ qp->resp_depth = attr->max_dest_rd_atomic;
}
qp_context->params2 |= cpu_to_be32(MTHCA_QP_BIT_RSC);
@@ -835,7 +879,7 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
}
err = mthca_MODIFY_QP(dev, state_table[cur_state][new_state].trans,
- qp->qpn, 0, qp_param, 0, &status);
+ qp->qpn, 0, mailbox, 0, &status);
if (status) {
mthca_warn(dev, "modify QP %d returned status %02x.\n",
state_table[cur_state][new_state].trans, status);
@@ -845,7 +889,7 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
if (!err)
qp->state = new_state;
- kfree(mailbox);
+ mthca_free_mailbox(dev, mailbox);
if (is_sqp(dev, qp))
store_attrs(to_msqp(qp), attr, attr_mask);
@@ -917,6 +961,15 @@ static int mthca_alloc_wqe_buf(struct mthca_dev *dev,
qp->send_wqe_offset = ALIGN(qp->rq.max << qp->rq.wqe_shift,
1 << qp->sq.wqe_shift);
+
+ /*
+ * If this is a userspace QP, we don't actually have to
+ * allocate anything. All we need is to calculate the WQE
+ * sizes and the send_wqe_offset, so we're done now.
+ */
+ if (pd->ibpd.uobject)
+ return 0;
+
size = PAGE_ALIGN(qp->send_wqe_offset +
(qp->sq.max << qp->sq.wqe_shift));
@@ -934,7 +987,8 @@ static int mthca_alloc_wqe_buf(struct mthca_dev *dev,
mthca_dbg(dev, "Creating direct QP of size %d (shift %d)\n",
size, shift);
- qp->queue.direct.buf = pci_alloc_consistent(dev->pdev, size, &t);
+ qp->queue.direct.buf = dma_alloc_coherent(&dev->pdev->dev, size,
+ &t, GFP_KERNEL);
if (!qp->queue.direct.buf)
goto err_out;
@@ -973,7 +1027,8 @@ static int mthca_alloc_wqe_buf(struct mthca_dev *dev,
for (i = 0; i < npages; ++i) {
qp->queue.page_list[i].buf =
- pci_alloc_consistent(dev->pdev, PAGE_SIZE, &t);
+ dma_alloc_coherent(&dev->pdev->dev, PAGE_SIZE,
+ &t, GFP_KERNEL);
if (!qp->queue.page_list[i].buf)
goto err_out_free;
@@ -996,16 +1051,15 @@ static int mthca_alloc_wqe_buf(struct mthca_dev *dev,
err_out_free:
if (qp->is_direct) {
- pci_free_consistent(dev->pdev, size,
- qp->queue.direct.buf,
- pci_unmap_addr(&qp->queue.direct, mapping));
+ dma_free_coherent(&dev->pdev->dev, size, qp->queue.direct.buf,
+ pci_unmap_addr(&qp->queue.direct, mapping));
} else
for (i = 0; i < npages; ++i) {
if (qp->queue.page_list[i].buf)
- pci_free_consistent(dev->pdev, PAGE_SIZE,
- qp->queue.page_list[i].buf,
- pci_unmap_addr(&qp->queue.page_list[i],
- mapping));
+ dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
+ qp->queue.page_list[i].buf,
+ pci_unmap_addr(&qp->queue.page_list[i],
+ mapping));
}
@@ -1015,10 +1069,32 @@ static int mthca_alloc_wqe_buf(struct mthca_dev *dev,
return err;
}
-static int mthca_alloc_memfree(struct mthca_dev *dev,
+static void mthca_free_wqe_buf(struct mthca_dev *dev,
struct mthca_qp *qp)
{
- int ret = 0;
+ int i;
+ int size = PAGE_ALIGN(qp->send_wqe_offset +
+ (qp->sq.max << qp->sq.wqe_shift));
+
+ if (qp->is_direct) {
+ dma_free_coherent(&dev->pdev->dev, size, qp->queue.direct.buf,
+ pci_unmap_addr(&qp->queue.direct, mapping));
+ } else {
+ for (i = 0; i < size / PAGE_SIZE; ++i) {
+ dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
+ qp->queue.page_list[i].buf,
+ pci_unmap_addr(&qp->queue.page_list[i],
+ mapping));
+ }
+ }
+
+ kfree(qp->wrid);
+}
+
+static int mthca_map_memfree(struct mthca_dev *dev,
+ struct mthca_qp *qp)
+{
+ int ret;
if (mthca_is_memfree(dev)) {
ret = mthca_table_get(dev, dev->qp_table.qp_table, qp->qpn);
@@ -1029,35 +1105,15 @@ static int mthca_alloc_memfree(struct mthca_dev *dev,
if (ret)
goto err_qpc;
- ret = mthca_table_get(dev, dev->qp_table.rdb_table,
- qp->qpn << dev->qp_table.rdb_shift);
- if (ret)
- goto err_eqpc;
-
- qp->rq.db_index = mthca_alloc_db(dev, MTHCA_DB_TYPE_RQ,
- qp->qpn, &qp->rq.db);
- if (qp->rq.db_index < 0) {
- ret = -ENOMEM;
- goto err_rdb;
- }
+ ret = mthca_table_get(dev, dev->qp_table.rdb_table,
+ qp->qpn << dev->qp_table.rdb_shift);
+ if (ret)
+ goto err_eqpc;
- qp->sq.db_index = mthca_alloc_db(dev, MTHCA_DB_TYPE_SQ,
- qp->qpn, &qp->sq.db);
- if (qp->sq.db_index < 0) {
- ret = -ENOMEM;
- goto err_rq_db;
- }
}
return 0;
-err_rq_db:
- mthca_free_db(dev, MTHCA_DB_TYPE_RQ, qp->rq.db_index);
-
-err_rdb:
- mthca_table_put(dev, dev->qp_table.rdb_table,
- qp->qpn << dev->qp_table.rdb_shift);
-
err_eqpc:
mthca_table_put(dev, dev->qp_table.eqp_table, qp->qpn);
@@ -1067,16 +1123,41 @@ err_qpc:
return ret;
}
+static void mthca_unmap_memfree(struct mthca_dev *dev,
+ struct mthca_qp *qp)
+{
+ mthca_table_put(dev, dev->qp_table.rdb_table,
+ qp->qpn << dev->qp_table.rdb_shift);
+ mthca_table_put(dev, dev->qp_table.eqp_table, qp->qpn);
+ mthca_table_put(dev, dev->qp_table.qp_table, qp->qpn);
+}
+
+static int mthca_alloc_memfree(struct mthca_dev *dev,
+ struct mthca_qp *qp)
+{
+ int ret = 0;
+
+ if (mthca_is_memfree(dev)) {
+ qp->rq.db_index = mthca_alloc_db(dev, MTHCA_DB_TYPE_RQ,
+ qp->qpn, &qp->rq.db);
+ if (qp->rq.db_index < 0)
+ return ret;
+
+ qp->sq.db_index = mthca_alloc_db(dev, MTHCA_DB_TYPE_SQ,
+ qp->qpn, &qp->sq.db);
+ if (qp->sq.db_index < 0)
+ mthca_free_db(dev, MTHCA_DB_TYPE_RQ, qp->rq.db_index);
+ }
+
+ return ret;
+}
+
static void mthca_free_memfree(struct mthca_dev *dev,
struct mthca_qp *qp)
{
if (mthca_is_memfree(dev)) {
mthca_free_db(dev, MTHCA_DB_TYPE_SQ, qp->sq.db_index);
mthca_free_db(dev, MTHCA_DB_TYPE_RQ, qp->rq.db_index);
- mthca_table_put(dev, dev->qp_table.rdb_table,
- qp->qpn << dev->qp_table.rdb_shift);
- mthca_table_put(dev, dev->qp_table.eqp_table, qp->qpn);
- mthca_table_put(dev, dev->qp_table.qp_table, qp->qpn);
}
}
@@ -1108,13 +1189,28 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
mthca_wq_init(&qp->sq);
mthca_wq_init(&qp->rq);
- ret = mthca_alloc_memfree(dev, qp);
+ ret = mthca_map_memfree(dev, qp);
if (ret)
return ret;
ret = mthca_alloc_wqe_buf(dev, pd, qp);
if (ret) {
- mthca_free_memfree(dev, qp);
+ mthca_unmap_memfree(dev, qp);
+ return ret;
+ }
+
+ /*
+ * If this is a userspace QP, we're done now. The doorbells
+ * will be allocated and buffers will be initialized in
+ * userspace.
+ */
+ if (pd->ibpd.uobject)
+ return 0;
+
+ ret = mthca_alloc_memfree(dev, qp);
+ if (ret) {
+ mthca_free_wqe_buf(dev, qp);
+ mthca_unmap_memfree(dev, qp);
return ret;
}
@@ -1147,22 +1243,39 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
return 0;
}
-static void mthca_align_qp_size(struct mthca_dev *dev, struct mthca_qp *qp)
+static int mthca_set_qp_size(struct mthca_dev *dev, struct ib_qp_cap *cap,
+ struct mthca_qp *qp)
{
- int i;
-
- if (!mthca_is_memfree(dev))
- return;
+ /* Sanity check QP size before proceeding */
+ if (cap->max_send_wr > 65536 || cap->max_recv_wr > 65536 ||
+ cap->max_send_sge > 64 || cap->max_recv_sge > 64)
+ return -EINVAL;
- for (i = 0; 1 << i < qp->rq.max; ++i)
- ; /* nothing */
+ if (mthca_is_memfree(dev)) {
+ qp->rq.max = cap->max_recv_wr ?
+ roundup_pow_of_two(cap->max_recv_wr) : 0;
+ qp->sq.max = cap->max_send_wr ?
+ roundup_pow_of_two(cap->max_send_wr) : 0;
+ } else {
+ qp->rq.max = cap->max_recv_wr;
+ qp->sq.max = cap->max_send_wr;
+ }
- qp->rq.max = 1 << i;
+ qp->rq.max_gs = cap->max_recv_sge;
+ qp->sq.max_gs = max_t(int, cap->max_send_sge,
+ ALIGN(cap->max_inline_data + MTHCA_INLINE_HEADER_SIZE,
+ MTHCA_INLINE_CHUNK_SIZE) /
+ sizeof (struct mthca_data_seg));
- for (i = 0; 1 << i < qp->sq.max; ++i)
- ; /* nothing */
+ /*
+ * For MLX transport we need 2 extra S/G entries:
+ * one for the header and one for the checksum at the end
+ */
+ if ((qp->transport == MLX && qp->sq.max_gs + 2 > dev->limits.max_sg) ||
+ qp->sq.max_gs > dev->limits.max_sg || qp->rq.max_gs > dev->limits.max_sg)
+ return -EINVAL;
- qp->sq.max = 1 << i;
+ return 0;
}
int mthca_alloc_qp(struct mthca_dev *dev,
@@ -1171,11 +1284,14 @@ int mthca_alloc_qp(struct mthca_dev *dev,
struct mthca_cq *recv_cq,
enum ib_qp_type type,
enum ib_sig_type send_policy,
+ struct ib_qp_cap *cap,
struct mthca_qp *qp)
{
int err;
- mthca_align_qp_size(dev, qp);
+ err = mthca_set_qp_size(dev, cap, qp);
+ if (err)
+ return err;
switch (type) {
case IB_QPT_RC: qp->transport = RC; break;
@@ -1208,14 +1324,17 @@ int mthca_alloc_sqp(struct mthca_dev *dev,
struct mthca_cq *send_cq,
struct mthca_cq *recv_cq,
enum ib_sig_type send_policy,
+ struct ib_qp_cap *cap,
int qpn,
int port,
struct mthca_sqp *sqp)
{
- int err = 0;
u32 mqpn = qpn * 2 + dev->qp_table.sqp_start + port - 1;
+ int err;
- mthca_align_qp_size(dev, &sqp->qp);
+ err = mthca_set_qp_size(dev, cap, &sqp->qp);
+ if (err)
+ return err;
sqp->header_buf_size = sqp->qp.sq.max * MTHCA_UD_HEADER_SIZE;
sqp->header_buf = dma_alloc_coherent(&dev->pdev->dev, sqp->header_buf_size,
@@ -1274,8 +1393,6 @@ void mthca_free_qp(struct mthca_dev *dev,
struct mthca_qp *qp)
{
u8 status;
- int size;
- int i;
struct mthca_cq *send_cq;
struct mthca_cq *recv_cq;
@@ -1305,31 +1422,22 @@ void mthca_free_qp(struct mthca_dev *dev,
if (qp->state != IB_QPS_RESET)
mthca_MODIFY_QP(dev, MTHCA_TRANS_ANY2RST, qp->qpn, 0, NULL, 0, &status);
- mthca_cq_clean(dev, to_mcq(qp->ibqp.send_cq)->cqn, qp->qpn);
- if (qp->ibqp.send_cq != qp->ibqp.recv_cq)
- mthca_cq_clean(dev, to_mcq(qp->ibqp.recv_cq)->cqn, qp->qpn);
-
- mthca_free_mr(dev, &qp->mr);
-
- size = PAGE_ALIGN(qp->send_wqe_offset +
- (qp->sq.max << qp->sq.wqe_shift));
+ /*
+ * If this is a userspace QP, the buffers, MR, CQs and so on
+ * will be cleaned up in userspace, so all we have to do is
+ * unref the mem-free tables and free the QPN in our table.
+ */
+ if (!qp->ibqp.uobject) {
+ mthca_cq_clean(dev, to_mcq(qp->ibqp.send_cq)->cqn, qp->qpn);
+ if (qp->ibqp.send_cq != qp->ibqp.recv_cq)
+ mthca_cq_clean(dev, to_mcq(qp->ibqp.recv_cq)->cqn, qp->qpn);
- if (qp->is_direct) {
- pci_free_consistent(dev->pdev, size,
- qp->queue.direct.buf,
- pci_unmap_addr(&qp->queue.direct, mapping));
- } else {
- for (i = 0; i < size / PAGE_SIZE; ++i) {
- pci_free_consistent(dev->pdev, PAGE_SIZE,
- qp->queue.page_list[i].buf,
- pci_unmap_addr(&qp->queue.page_list[i],
- mapping));
- }
+ mthca_free_mr(dev, &qp->mr);
+ mthca_free_memfree(dev, qp);
+ mthca_free_wqe_buf(dev, qp);
}
- kfree(qp->wrid);
-
- mthca_free_memfree(dev, qp);
+ mthca_unmap_memfree(dev, qp);
if (is_sqp(dev, qp)) {
atomic_dec(&(to_mpd(qp->ibqp.pd)->sqp_count));
@@ -1529,6 +1637,26 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
break;
+ case UC:
+ switch (wr->opcode) {
+ case IB_WR_RDMA_WRITE:
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ ((struct mthca_raddr_seg *) wqe)->raddr =
+ cpu_to_be64(wr->wr.rdma.remote_addr);
+ ((struct mthca_raddr_seg *) wqe)->rkey =
+ cpu_to_be32(wr->wr.rdma.rkey);
+ ((struct mthca_raddr_seg *) wqe)->reserved = 0;
+ wqe += sizeof (struct mthca_raddr_seg);
+ size += sizeof (struct mthca_raddr_seg) / 16;
+ break;
+
+ default:
+ /* No extra segments required for sends */
+ break;
+ }
+
+ break;
+
case UD:
((struct mthca_tavor_ud_seg *) wqe)->lkey =
cpu_to_be32(to_mah(wr->wr.ud.ah)->key);
@@ -1814,9 +1942,29 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
sizeof (struct mthca_atomic_seg);
break;
+ case IB_WR_RDMA_READ:
+ case IB_WR_RDMA_WRITE:
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ ((struct mthca_raddr_seg *) wqe)->raddr =
+ cpu_to_be64(wr->wr.rdma.remote_addr);
+ ((struct mthca_raddr_seg *) wqe)->rkey =
+ cpu_to_be32(wr->wr.rdma.rkey);
+ ((struct mthca_raddr_seg *) wqe)->reserved = 0;
+ wqe += sizeof (struct mthca_raddr_seg);
+ size += sizeof (struct mthca_raddr_seg) / 16;
+ break;
+
+ default:
+ /* No extra segments required for sends */
+ break;
+ }
+
+ break;
+
+ case UC:
+ switch (wr->opcode) {
case IB_WR_RDMA_WRITE:
case IB_WR_RDMA_WRITE_WITH_IMM:
- case IB_WR_RDMA_READ:
((struct mthca_raddr_seg *) wqe)->raddr =
cpu_to_be64(wr->wr.rdma.remote_addr);
((struct mthca_raddr_seg *) wqe)->rkey =