int mthca_state;
u8 status;
+ mutex_lock(&qp->mutex);
+
if (qp->state == IB_QPS_RESET) {
qp_attr->qp_state = IB_QPS_RESET;
goto done;
}
mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
- if (IS_ERR(mailbox))
- return PTR_ERR(mailbox);
+ if (IS_ERR(mailbox)) {
+ err = PTR_ERR(mailbox);
+ goto out;
+ }
err = mthca_QUERY_QP(dev, qp->qpn, 0, mailbox, &status);
if (err)
- goto out;
+ goto out_mailbox;
if (status) {
mthca_warn(dev, "QUERY_QP returned status %02x\n", status);
err = -EINVAL;
- goto out;
+ goto out_mailbox;
}
qp_param = mailbox->buf;
context = &qp_param->context;
mthca_state = be32_to_cpu(context->flags) >> 28;
- qp_attr->qp_state = to_ib_qp_state(mthca_state);
+ qp->state = to_ib_qp_state(mthca_state);
+ qp_attr->qp_state = qp->state;
qp_attr->path_mtu = context->mtu_msgmax >> 5;
qp_attr->path_mig_state =
to_ib_mig_state((be32_to_cpu(context->flags) >> 11) & 0x3);
qp_init_attr->cap = qp_attr->cap;
-out:
+out_mailbox:
mthca_free_mailbox(dev, mailbox);
+
+out:
+ mutex_unlock(&qp->mutex);
return err;
}
{
int ret;
int i;
+ struct mthca_next_seg *next;
qp->refcount = 1;
init_waitqueue_head(&qp->wait);
}
if (mthca_is_memfree(dev)) {
- struct mthca_next_seg *next;
struct mthca_data_seg *scatter;
int size = (sizeof (struct mthca_next_seg) +
qp->rq.max_gs * sizeof (struct mthca_data_seg)) / 16;
qp->sq.wqe_shift) +
qp->send_wqe_offset);
}
+ } else {
+ for (i = 0; i < qp->rq.max; ++i) {
+ next = get_recv_wqe(qp, i);
+ next->nda_op = htonl((((i + 1) % qp->rq.max) <<
+ qp->rq.wqe_shift) | 1);
+ }
+
}
qp->sq.last = get_send_wqe(qp, qp->sq.max - 1);
case IB_WR_SEND_WITH_IMM:
sqp->ud_header.bth.opcode = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE;
sqp->ud_header.immediate_present = 1;
- sqp->ud_header.immediate_data = wr->imm_data;
+ sqp->ud_header.immediate_data = wr->ex.imm_data;
break;
default:
return -EINVAL;
return cur + nreq >= wq->max;
}
+static __always_inline void set_raddr_seg(struct mthca_raddr_seg *rseg,
+ u64 remote_addr, u32 rkey)
+{
+ rseg->raddr = cpu_to_be64(remote_addr);
+ rseg->rkey = cpu_to_be32(rkey);
+ rseg->reserved = 0;
+}
+
+static __always_inline void set_atomic_seg(struct mthca_atomic_seg *aseg,
+ struct ib_send_wr *wr)
+{
+ if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
+ aseg->swap_add = cpu_to_be64(wr->wr.atomic.swap);
+ aseg->compare = cpu_to_be64(wr->wr.atomic.compare_add);
+ } else {
+ aseg->swap_add = cpu_to_be64(wr->wr.atomic.compare_add);
+ aseg->compare = 0;
+ }
+
+}
+
+static void set_tavor_ud_seg(struct mthca_tavor_ud_seg *useg,
+ struct ib_send_wr *wr)
+{
+ useg->lkey = cpu_to_be32(to_mah(wr->wr.ud.ah)->key);
+ useg->av_addr = cpu_to_be64(to_mah(wr->wr.ud.ah)->avdma);
+ useg->dqpn = cpu_to_be32(wr->wr.ud.remote_qpn);
+ useg->qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+
+}
+
+static void set_arbel_ud_seg(struct mthca_arbel_ud_seg *useg,
+ struct ib_send_wr *wr)
+{
+ memcpy(useg->av, to_mah(wr->wr.ud.ah)->av, MTHCA_AV_SIZE);
+ useg->dqpn = cpu_to_be32(wr->wr.ud.remote_qpn);
+ useg->qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+}
+
int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
struct ib_send_wr **bad_wr)
{
int nreq;
int i;
int size;
- int size0 = 0;
- u32 f0 = 0;
+ /*
+ * f0 and size0 are only used if nreq != 0, and they will
+ * always be initialized the first time through the main loop
+ * before nreq is incremented. So nreq cannot become non-zero
+ * without initializing f0 and size0, and they are in fact
+ * never used uninitialized.
+ */
+ int uninitialized_var(size0);
+ u32 uninitialized_var(f0);
int ind;
u8 op0 = 0;
cpu_to_be32(1);
if (wr->opcode == IB_WR_SEND_WITH_IMM ||
wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM)
- ((struct mthca_next_seg *) wqe)->imm = wr->imm_data;
+ ((struct mthca_next_seg *) wqe)->imm = wr->ex.imm_data;
wqe += sizeof (struct mthca_next_seg);
size = sizeof (struct mthca_next_seg) / 16;
switch (wr->opcode) {
case IB_WR_ATOMIC_CMP_AND_SWP:
case IB_WR_ATOMIC_FETCH_AND_ADD:
- ((struct mthca_raddr_seg *) wqe)->raddr =
- cpu_to_be64(wr->wr.atomic.remote_addr);
- ((struct mthca_raddr_seg *) wqe)->rkey =
- cpu_to_be32(wr->wr.atomic.rkey);
- ((struct mthca_raddr_seg *) wqe)->reserved = 0;
-
+ set_raddr_seg(wqe, wr->wr.atomic.remote_addr,
+ wr->wr.atomic.rkey);
wqe += sizeof (struct mthca_raddr_seg);
- if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
- ((struct mthca_atomic_seg *) wqe)->swap_add =
- cpu_to_be64(wr->wr.atomic.swap);
- ((struct mthca_atomic_seg *) wqe)->compare =
- cpu_to_be64(wr->wr.atomic.compare_add);
- } else {
- ((struct mthca_atomic_seg *) wqe)->swap_add =
- cpu_to_be64(wr->wr.atomic.compare_add);
- ((struct mthca_atomic_seg *) wqe)->compare = 0;
- }
-
+ set_atomic_seg(wqe, wr);
wqe += sizeof (struct mthca_atomic_seg);
size += (sizeof (struct mthca_raddr_seg) +
sizeof (struct mthca_atomic_seg)) / 16;
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 =
- cpu_to_be32(wr->wr.rdma.rkey);
- ((struct mthca_raddr_seg *) wqe)->reserved = 0;
- wqe += sizeof (struct mthca_raddr_seg);
+ set_raddr_seg(wqe, wr->wr.rdma.remote_addr,
+ wr->wr.rdma.rkey);
+ wqe += sizeof (struct mthca_raddr_seg);
size += sizeof (struct mthca_raddr_seg) / 16;
break;
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);
+ set_raddr_seg(wqe, wr->wr.rdma.remote_addr,
+ wr->wr.rdma.rkey);
+ wqe += sizeof (struct mthca_raddr_seg);
size += sizeof (struct mthca_raddr_seg) / 16;
break;
break;
case UD:
- ((struct mthca_tavor_ud_seg *) wqe)->lkey =
- cpu_to_be32(to_mah(wr->wr.ud.ah)->key);
- ((struct mthca_tavor_ud_seg *) wqe)->av_addr =
- cpu_to_be64(to_mah(wr->wr.ud.ah)->avdma);
- ((struct mthca_tavor_ud_seg *) wqe)->dqpn =
- cpu_to_be32(wr->wr.ud.remote_qpn);
- ((struct mthca_tavor_ud_seg *) wqe)->qkey =
- cpu_to_be32(wr->wr.ud.remote_qkey);
-
- wqe += sizeof (struct mthca_tavor_ud_seg);
+ set_tavor_ud_seg(wqe, wr);
+ wqe += sizeof (struct mthca_tavor_ud_seg);
size += sizeof (struct mthca_tavor_ud_seg) / 16;
break;
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
- wqe += sizeof (struct mthca_data_seg);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
+ wqe += sizeof (struct mthca_data_seg);
size += sizeof (struct mthca_data_seg) / 16;
}
mthca_opcode[wr->opcode]);
wmb();
((struct mthca_next_seg *) prev_wqe)->ee_nds =
- cpu_to_be32((size0 ? 0 : MTHCA_NEXT_DBD) | size |
+ cpu_to_be32((nreq ? 0 : MTHCA_NEXT_DBD) | size |
((wr->send_flags & IB_SEND_FENCE) ?
MTHCA_NEXT_FENCE : 0));
- if (!size0) {
+ if (!nreq) {
size0 = size;
op0 = mthca_opcode[wr->opcode];
f0 = wr->send_flags & IB_SEND_FENCE ?
out:
if (likely(nreq)) {
- __be32 doorbell[2];
-
- doorbell[0] = cpu_to_be32(((qp->sq.next_ind << qp->sq.wqe_shift) +
- qp->send_wqe_offset) | f0 | op0);
- doorbell[1] = cpu_to_be32((qp->qpn << 8) | size0);
-
wmb();
- mthca_write64(doorbell,
+ mthca_write64(((qp->sq.next_ind << qp->sq.wqe_shift) +
+ qp->send_wqe_offset) | f0 | op0,
+ (qp->qpn << 8) | size0,
dev->kar + MTHCA_SEND_DOORBELL,
MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
/*
{
struct mthca_dev *dev = to_mdev(ibqp->device);
struct mthca_qp *qp = to_mqp(ibqp);
- __be32 doorbell[2];
unsigned long flags;
int err = 0;
int nreq;
int i;
int size;
- int size0 = 0;
+ /*
+ * size0 is only used if nreq != 0, and it will always be
+ * initialized the first time through the main loop before
+ * nreq is incremented. So nreq cannot become non-zero
+ * without initializing size0, and it is in fact never used
+ * uninitialized.
+ */
+ int uninitialized_var(size0);
int ind;
void *wqe;
void *prev_wqe;
prev_wqe = qp->rq.last;
qp->rq.last = wqe;
- ((struct mthca_next_seg *) wqe)->nda_op = 0;
((struct mthca_next_seg *) wqe)->ee_nds =
cpu_to_be32(MTHCA_NEXT_DBD);
((struct mthca_next_seg *) wqe)->flags = 0;
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
- wqe += sizeof (struct mthca_data_seg);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
+ wqe += sizeof (struct mthca_data_seg);
size += sizeof (struct mthca_data_seg) / 16;
}
qp->wrid[ind] = wr->wr_id;
- ((struct mthca_next_seg *) prev_wqe)->nda_op =
- cpu_to_be32((ind << qp->rq.wqe_shift) | 1);
- wmb();
((struct mthca_next_seg *) prev_wqe)->ee_nds =
cpu_to_be32(MTHCA_NEXT_DBD | size);
- if (!size0)
+ if (!nreq)
size0 = size;
++ind;
if (unlikely(nreq == MTHCA_TAVOR_MAX_WQES_PER_RECV_DB)) {
nreq = 0;
- doorbell[0] = cpu_to_be32((qp->rq.next_ind << qp->rq.wqe_shift) | size0);
- doorbell[1] = cpu_to_be32(qp->qpn << 8);
-
wmb();
- mthca_write64(doorbell,
- dev->kar + MTHCA_RECEIVE_DOORBELL,
+ mthca_write64((qp->rq.next_ind << qp->rq.wqe_shift) | size0,
+ qp->qpn << 8, dev->kar + MTHCA_RECEIVE_DOORBELL,
MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
qp->rq.next_ind = ind;
qp->rq.head += MTHCA_TAVOR_MAX_WQES_PER_RECV_DB;
- size0 = 0;
}
}
out:
if (likely(nreq)) {
- doorbell[0] = cpu_to_be32((qp->rq.next_ind << qp->rq.wqe_shift) | size0);
- doorbell[1] = cpu_to_be32((qp->qpn << 8) | nreq);
-
wmb();
- mthca_write64(doorbell,
- dev->kar + MTHCA_RECEIVE_DOORBELL,
+ mthca_write64((qp->rq.next_ind << qp->rq.wqe_shift) | size0,
+ qp->qpn << 8 | nreq, dev->kar + MTHCA_RECEIVE_DOORBELL,
MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
}
{
struct mthca_dev *dev = to_mdev(ibqp->device);
struct mthca_qp *qp = to_mqp(ibqp);
- __be32 doorbell[2];
+ u32 dbhi;
void *wqe;
void *prev_wqe;
unsigned long flags;
int nreq;
int i;
int size;
- int size0 = 0;
- u32 f0 = 0;
+ /*
+ * f0 and size0 are only used if nreq != 0, and they will
+ * always be initialized the first time through the main loop
+ * before nreq is incremented. So nreq cannot become non-zero
+ * without initializing f0 and size0, and they are in fact
+ * never used uninitialized.
+ */
+ int uninitialized_var(size0);
+ u32 uninitialized_var(f0);
int ind;
u8 op0 = 0;
if (unlikely(nreq == MTHCA_ARBEL_MAX_WQES_PER_SEND_DB)) {
nreq = 0;
- doorbell[0] = cpu_to_be32((MTHCA_ARBEL_MAX_WQES_PER_SEND_DB << 24) |
- ((qp->sq.head & 0xffff) << 8) |
- f0 | op0);
- doorbell[1] = cpu_to_be32((qp->qpn << 8) | size0);
+ dbhi = (MTHCA_ARBEL_MAX_WQES_PER_SEND_DB << 24) |
+ ((qp->sq.head & 0xffff) << 8) | f0 | op0;
qp->sq.head += MTHCA_ARBEL_MAX_WQES_PER_SEND_DB;
- size0 = 0;
/*
* Make sure that descriptors are written before
* write MMIO send doorbell.
*/
wmb();
- mthca_write64(doorbell,
+
+ mthca_write64(dbhi, (qp->qpn << 8) | size0,
dev->kar + MTHCA_SEND_DOORBELL,
MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
}
cpu_to_be32(MTHCA_NEXT_CQ_UPDATE) : 0) |
((wr->send_flags & IB_SEND_SOLICITED) ?
cpu_to_be32(MTHCA_NEXT_SOLICIT) : 0) |
+ ((wr->send_flags & IB_SEND_IP_CSUM) ?
+ cpu_to_be32(MTHCA_NEXT_IP_CSUM | MTHCA_NEXT_TCP_UDP_CSUM) : 0) |
cpu_to_be32(1);
if (wr->opcode == IB_WR_SEND_WITH_IMM ||
wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM)
- ((struct mthca_next_seg *) wqe)->imm = wr->imm_data;
+ ((struct mthca_next_seg *) wqe)->imm = wr->ex.imm_data;
wqe += sizeof (struct mthca_next_seg);
size = sizeof (struct mthca_next_seg) / 16;
switch (wr->opcode) {
case IB_WR_ATOMIC_CMP_AND_SWP:
case IB_WR_ATOMIC_FETCH_AND_ADD:
- ((struct mthca_raddr_seg *) wqe)->raddr =
- cpu_to_be64(wr->wr.atomic.remote_addr);
- ((struct mthca_raddr_seg *) wqe)->rkey =
- cpu_to_be32(wr->wr.atomic.rkey);
- ((struct mthca_raddr_seg *) wqe)->reserved = 0;
-
+ set_raddr_seg(wqe, wr->wr.atomic.remote_addr,
+ wr->wr.atomic.rkey);
wqe += sizeof (struct mthca_raddr_seg);
- if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
- ((struct mthca_atomic_seg *) wqe)->swap_add =
- cpu_to_be64(wr->wr.atomic.swap);
- ((struct mthca_atomic_seg *) wqe)->compare =
- cpu_to_be64(wr->wr.atomic.compare_add);
- } else {
- ((struct mthca_atomic_seg *) wqe)->swap_add =
- cpu_to_be64(wr->wr.atomic.compare_add);
- ((struct mthca_atomic_seg *) wqe)->compare = 0;
- }
-
- wqe += sizeof (struct mthca_atomic_seg);
+ set_atomic_seg(wqe, wr);
+ wqe += sizeof (struct mthca_atomic_seg);
size += (sizeof (struct mthca_raddr_seg) +
sizeof (struct mthca_atomic_seg)) / 16;
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);
+ set_raddr_seg(wqe, wr->wr.rdma.remote_addr,
+ wr->wr.rdma.rkey);
+ wqe += sizeof (struct mthca_raddr_seg);
size += sizeof (struct mthca_raddr_seg) / 16;
break;
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);
+ set_raddr_seg(wqe, wr->wr.rdma.remote_addr,
+ wr->wr.rdma.rkey);
+ wqe += sizeof (struct mthca_raddr_seg);
size += sizeof (struct mthca_raddr_seg) / 16;
break;
break;
case UD:
- memcpy(((struct mthca_arbel_ud_seg *) wqe)->av,
- to_mah(wr->wr.ud.ah)->av, MTHCA_AV_SIZE);
- ((struct mthca_arbel_ud_seg *) wqe)->dqpn =
- cpu_to_be32(wr->wr.ud.remote_qpn);
- ((struct mthca_arbel_ud_seg *) wqe)->qkey =
- cpu_to_be32(wr->wr.ud.remote_qkey);
-
- wqe += sizeof (struct mthca_arbel_ud_seg);
+ set_arbel_ud_seg(wqe, wr);
+ wqe += sizeof (struct mthca_arbel_ud_seg);
size += sizeof (struct mthca_arbel_ud_seg) / 16;
break;
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
- wqe += sizeof (struct mthca_data_seg);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
+ wqe += sizeof (struct mthca_data_seg);
size += sizeof (struct mthca_data_seg) / 16;
}
((wr->send_flags & IB_SEND_FENCE) ?
MTHCA_NEXT_FENCE : 0));
- if (!size0) {
+ if (!nreq) {
size0 = size;
op0 = mthca_opcode[wr->opcode];
f0 = wr->send_flags & IB_SEND_FENCE ?
out:
if (likely(nreq)) {
- doorbell[0] = cpu_to_be32((nreq << 24) |
- ((qp->sq.head & 0xffff) << 8) |
- f0 | op0);
- doorbell[1] = cpu_to_be32((qp->qpn << 8) | size0);
+ dbhi = (nreq << 24) | ((qp->sq.head & 0xffff) << 8) | f0 | op0;
qp->sq.head += nreq;
* write MMIO send doorbell.
*/
wmb();
- mthca_write64(doorbell,
- dev->kar + MTHCA_SEND_DOORBELL,
+
+ mthca_write64(dbhi, (qp->qpn << 8) | size0, dev->kar + MTHCA_SEND_DOORBELL,
MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
}
}
for (i = 0; i < wr->num_sge; ++i) {
- ((struct mthca_data_seg *) wqe)->byte_count =
- cpu_to_be32(wr->sg_list[i].length);
- ((struct mthca_data_seg *) wqe)->lkey =
- cpu_to_be32(wr->sg_list[i].lkey);
- ((struct mthca_data_seg *) wqe)->addr =
- cpu_to_be64(wr->sg_list[i].addr);
+ mthca_set_data_seg(wqe, wr->sg_list + i);
wqe += sizeof (struct mthca_data_seg);
}
- if (i < qp->rq.max_gs) {
- ((struct mthca_data_seg *) wqe)->byte_count = 0;
- ((struct mthca_data_seg *) wqe)->lkey = cpu_to_be32(MTHCA_INVAL_LKEY);
- ((struct mthca_data_seg *) wqe)->addr = 0;
- }
+ if (i < qp->rq.max_gs)
+ mthca_set_data_seg_inval(wqe);
qp->wrid[ind] = wr->wr_id;