summaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/hw/ipath/ipath_ruc.c
diff options
context:
space:
mode:
authorBryan O'Sullivan <bos@pathscale.com>2006-07-01 04:35:50 -0700
committerLinus Torvalds <torvalds@g5.osdl.org>2006-07-01 09:55:58 -0700
commitddd4bb22108417fdc5c35324bd13a3265581ae76 (patch)
tree5cc35e54e8761af27746bae48ef66318237ad8a0 /drivers/infiniband/hw/ipath/ipath_ruc.c
parent759d57686dab8169ca68bbf938ce8e965d1e107a (diff)
downloadop-kernel-dev-ddd4bb22108417fdc5c35324bd13a3265581ae76.zip
op-kernel-dev-ddd4bb22108417fdc5c35324bd13a3265581ae76.tar.gz
[PATCH] IB/ipath: share more common code between RC and UC protocols
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com> Signed-off-by: Bryan O'Sullivan <bryan.osullivan@qlogic.com> Cc: "Michael S. Tsirkin" <mst@mellanox.co.il> Cc: Roland Dreier <rolandd@cisco.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'drivers/infiniband/hw/ipath/ipath_ruc.c')
-rw-r--r--drivers/infiniband/hw/ipath/ipath_ruc.c245
1 files changed, 191 insertions, 54 deletions
diff --git a/drivers/infiniband/hw/ipath/ipath_ruc.c b/drivers/infiniband/hw/ipath/ipath_ruc.c
index 2714d64..9a456a7 100644
--- a/drivers/infiniband/hw/ipath/ipath_ruc.c
+++ b/drivers/infiniband/hw/ipath/ipath_ruc.c
@@ -32,6 +32,7 @@
*/
#include "ipath_verbs.h"
+#include "ips_common.h"
/*
* Convert the AETH RNR timeout code into the number of milliseconds.
@@ -188,7 +189,6 @@ bail:
/**
* ipath_ruc_loopback - handle UC and RC lookback requests
* @sqp: the loopback QP
- * @wc: the work completion entry
*
* This is called from ipath_do_uc_send() or ipath_do_rc_send() to
* forward a WQE addressed to the same HCA.
@@ -197,13 +197,14 @@ bail:
* receive interrupts since this is a connected protocol and all packets
* will pass through here.
*/
-void ipath_ruc_loopback(struct ipath_qp *sqp, struct ib_wc *wc)
+static void ipath_ruc_loopback(struct ipath_qp *sqp)
{
struct ipath_ibdev *dev = to_idev(sqp->ibqp.device);
struct ipath_qp *qp;
struct ipath_swqe *wqe;
struct ipath_sge *sge;
unsigned long flags;
+ struct ib_wc wc;
u64 sdata;
qp = ipath_lookup_qpn(&dev->qp_table, sqp->remote_qpn);
@@ -234,8 +235,8 @@ again:
wqe = get_swqe_ptr(sqp, sqp->s_last);
spin_unlock_irqrestore(&sqp->s_lock, flags);
- wc->wc_flags = 0;
- wc->imm_data = 0;
+ wc.wc_flags = 0;
+ wc.imm_data = 0;
sqp->s_sge.sge = wqe->sg_list[0];
sqp->s_sge.sg_list = wqe->sg_list + 1;
@@ -243,8 +244,8 @@ again:
sqp->s_len = wqe->length;
switch (wqe->wr.opcode) {
case IB_WR_SEND_WITH_IMM:
- wc->wc_flags = IB_WC_WITH_IMM;
- wc->imm_data = wqe->wr.imm_data;
+ wc.wc_flags = IB_WC_WITH_IMM;
+ wc.imm_data = wqe->wr.imm_data;
/* FALLTHROUGH */
case IB_WR_SEND:
spin_lock_irqsave(&qp->r_rq.lock, flags);
@@ -255,7 +256,7 @@ again:
if (qp->ibqp.qp_type == IB_QPT_UC)
goto send_comp;
if (sqp->s_rnr_retry == 0) {
- wc->status = IB_WC_RNR_RETRY_EXC_ERR;
+ wc.status = IB_WC_RNR_RETRY_EXC_ERR;
goto err;
}
if (sqp->s_rnr_retry_cnt < 7)
@@ -270,8 +271,8 @@ again:
break;
case IB_WR_RDMA_WRITE_WITH_IMM:
- wc->wc_flags = IB_WC_WITH_IMM;
- wc->imm_data = wqe->wr.imm_data;
+ wc.wc_flags = IB_WC_WITH_IMM;
+ wc.imm_data = wqe->wr.imm_data;
spin_lock_irqsave(&qp->r_rq.lock, flags);
if (!ipath_get_rwqe(qp, 1))
goto rnr_nak;
@@ -285,20 +286,20 @@ again:
wqe->wr.wr.rdma.rkey,
IB_ACCESS_REMOTE_WRITE))) {
acc_err:
- wc->status = IB_WC_REM_ACCESS_ERR;
+ wc.status = IB_WC_REM_ACCESS_ERR;
err:
- wc->wr_id = wqe->wr.wr_id;
- wc->opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
- wc->vendor_err = 0;
- wc->byte_len = 0;
- wc->qp_num = sqp->ibqp.qp_num;
- wc->src_qp = sqp->remote_qpn;
- wc->pkey_index = 0;
- wc->slid = sqp->remote_ah_attr.dlid;
- wc->sl = sqp->remote_ah_attr.sl;
- wc->dlid_path_bits = 0;
- wc->port_num = 0;
- ipath_sqerror_qp(sqp, wc);
+ wc.wr_id = wqe->wr.wr_id;
+ wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
+ wc.vendor_err = 0;
+ wc.byte_len = 0;
+ wc.qp_num = sqp->ibqp.qp_num;
+ wc.src_qp = sqp->remote_qpn;
+ wc.pkey_index = 0;
+ wc.slid = sqp->remote_ah_attr.dlid;
+ wc.sl = sqp->remote_ah_attr.sl;
+ wc.dlid_path_bits = 0;
+ wc.port_num = 0;
+ ipath_sqerror_qp(sqp, &wc);
goto done;
}
break;
@@ -374,22 +375,22 @@ again:
goto send_comp;
if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM)
- wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
+ wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
else
- wc->opcode = IB_WC_RECV;
- wc->wr_id = qp->r_wr_id;
- wc->status = IB_WC_SUCCESS;
- wc->vendor_err = 0;
- wc->byte_len = wqe->length;
- wc->qp_num = qp->ibqp.qp_num;
- wc->src_qp = qp->remote_qpn;
+ wc.opcode = IB_WC_RECV;
+ wc.wr_id = qp->r_wr_id;
+ wc.status = IB_WC_SUCCESS;
+ wc.vendor_err = 0;
+ wc.byte_len = wqe->length;
+ wc.qp_num = qp->ibqp.qp_num;
+ wc.src_qp = qp->remote_qpn;
/* XXX do we know which pkey matched? Only needed for GSI. */
- wc->pkey_index = 0;
- wc->slid = qp->remote_ah_attr.dlid;
- wc->sl = qp->remote_ah_attr.sl;
- wc->dlid_path_bits = 0;
+ wc.pkey_index = 0;
+ wc.slid = qp->remote_ah_attr.dlid;
+ wc.sl = qp->remote_ah_attr.sl;
+ wc.dlid_path_bits = 0;
/* Signal completion event if the solicited bit is set. */
- ipath_cq_enter(to_icq(qp->ibqp.recv_cq), wc,
+ ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
wqe->wr.send_flags & IB_SEND_SOLICITED);
send_comp:
@@ -397,19 +398,19 @@ send_comp:
if (!test_bit(IPATH_S_SIGNAL_REQ_WR, &sqp->s_flags) ||
(wqe->wr.send_flags & IB_SEND_SIGNALED)) {
- wc->wr_id = wqe->wr.wr_id;
- wc->status = IB_WC_SUCCESS;
- wc->opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
- wc->vendor_err = 0;
- wc->byte_len = wqe->length;
- wc->qp_num = sqp->ibqp.qp_num;
- wc->src_qp = 0;
- wc->pkey_index = 0;
- wc->slid = 0;
- wc->sl = 0;
- wc->dlid_path_bits = 0;
- wc->port_num = 0;
- ipath_cq_enter(to_icq(sqp->ibqp.send_cq), wc, 0);
+ wc.wr_id = wqe->wr.wr_id;
+ wc.status = IB_WC_SUCCESS;
+ wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
+ wc.vendor_err = 0;
+ wc.byte_len = wqe->length;
+ wc.qp_num = sqp->ibqp.qp_num;
+ wc.src_qp = 0;
+ wc.pkey_index = 0;
+ wc.slid = 0;
+ wc.sl = 0;
+ wc.dlid_path_bits = 0;
+ wc.port_num = 0;
+ ipath_cq_enter(to_icq(sqp->ibqp.send_cq), &wc, 0);
}
/* Update s_last now that we are finished with the SWQE */
@@ -455,11 +456,11 @@ void ipath_no_bufs_available(struct ipath_qp *qp, struct ipath_ibdev *dev)
}
/**
- * ipath_post_rc_send - post RC and UC sends
+ * ipath_post_ruc_send - post RC and UC sends
* @qp: the QP to post on
* @wr: the work request to send
*/
-int ipath_post_rc_send(struct ipath_qp *qp, struct ib_send_wr *wr)
+int ipath_post_ruc_send(struct ipath_qp *qp, struct ib_send_wr *wr)
{
struct ipath_swqe *wqe;
unsigned long flags;
@@ -534,13 +535,149 @@ int ipath_post_rc_send(struct ipath_qp *qp, struct ib_send_wr *wr)
qp->s_head = next;
spin_unlock_irqrestore(&qp->s_lock, flags);
- if (qp->ibqp.qp_type == IB_QPT_UC)
- ipath_do_uc_send((unsigned long) qp);
- else
- ipath_do_rc_send((unsigned long) qp);
+ ipath_do_ruc_send((unsigned long) qp);
ret = 0;
bail:
return ret;
}
+
+/**
+ * ipath_make_grh - construct a GRH header
+ * @dev: a pointer to the ipath device
+ * @hdr: a pointer to the GRH header being constructed
+ * @grh: the global route address to send to
+ * @hwords: the number of 32 bit words of header being sent
+ * @nwords: the number of 32 bit words of data being sent
+ *
+ * Return the size of the header in 32 bit words.
+ */
+u32 ipath_make_grh(struct ipath_ibdev *dev, struct ib_grh *hdr,
+ struct ib_global_route *grh, u32 hwords, u32 nwords)
+{
+ hdr->version_tclass_flow =
+ cpu_to_be32((6 << 28) |
+ (grh->traffic_class << 20) |
+ grh->flow_label);
+ hdr->paylen = cpu_to_be16((hwords - 2 + nwords + SIZE_OF_CRC) << 2);
+ /* next_hdr is defined by C8-7 in ch. 8.4.1 */
+ hdr->next_hdr = 0x1B;
+ hdr->hop_limit = grh->hop_limit;
+ /* The SGID is 32-bit aligned. */
+ hdr->sgid.global.subnet_prefix = dev->gid_prefix;
+ hdr->sgid.global.interface_id = ipath_layer_get_guid(dev->dd);
+ hdr->dgid = grh->dgid;
+
+ /* GRH header size in 32-bit words. */
+ return sizeof(struct ib_grh) / sizeof(u32);
+}
+
+/**
+ * ipath_do_ruc_send - perform a send on an RC or UC QP
+ * @data: contains a pointer to the QP
+ *
+ * Process entries in the send work queue until credit or queue is
+ * exhausted. Only allow one CPU to send a packet per QP (tasklet).
+ * Otherwise, after we drop the QP s_lock, two threads could send
+ * packets out of order.
+ */
+void ipath_do_ruc_send(unsigned long data)
+{
+ struct ipath_qp *qp = (struct ipath_qp *)data;
+ struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
+ unsigned long flags;
+ u16 lrh0;
+ u32 nwords;
+ u32 extra_bytes;
+ u32 bth0;
+ u32 bth2;
+ u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+ struct ipath_other_headers *ohdr;
+
+ if (test_and_set_bit(IPATH_S_BUSY, &qp->s_flags))
+ goto bail;
+
+ if (unlikely(qp->remote_ah_attr.dlid ==
+ ipath_layer_get_lid(dev->dd))) {
+ ipath_ruc_loopback(qp);
+ goto clear;
+ }
+
+ ohdr = &qp->s_hdr.u.oth;
+ if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+ ohdr = &qp->s_hdr.u.l.oth;
+
+again:
+ /* Check for a constructed packet to be sent. */
+ if (qp->s_hdrwords != 0) {
+ /*
+ * If no PIO bufs are available, return. An interrupt will
+ * call ipath_ib_piobufavail() when one is available.
+ */
+ if (ipath_verbs_send(dev->dd, qp->s_hdrwords,
+ (u32 *) &qp->s_hdr, qp->s_cur_size,
+ qp->s_cur_sge)) {
+ ipath_no_bufs_available(qp, dev);
+ goto bail;
+ }
+ dev->n_unicast_xmit++;
+ /* Record that we sent the packet and s_hdr is empty. */
+ qp->s_hdrwords = 0;
+ }
+
+ /*
+ * The lock is needed to synchronize between setting
+ * qp->s_ack_state, resend timer, and post_send().
+ */
+ spin_lock_irqsave(&qp->s_lock, flags);
+
+ /* Sending responses has higher priority over sending requests. */
+ if (qp->s_ack_state != IB_OPCODE_RC_ACKNOWLEDGE &&
+ (bth0 = ipath_make_rc_ack(qp, ohdr, pmtu)) != 0)
+ bth2 = qp->s_ack_psn++ & IPS_PSN_MASK;
+ else if (!((qp->ibqp.qp_type == IB_QPT_RC) ?
+ ipath_make_rc_req(qp, ohdr, pmtu, &bth0, &bth2) :
+ ipath_make_uc_req(qp, ohdr, pmtu, &bth0, &bth2))) {
+ /*
+ * Clear the busy bit before unlocking to avoid races with
+ * adding new work queue items and then failing to process
+ * them.
+ */
+ clear_bit(IPATH_S_BUSY, &qp->s_flags);
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+ goto bail;
+ }
+
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+
+ /* Construct the header. */
+ extra_bytes = (4 - qp->s_cur_size) & 3;
+ nwords = (qp->s_cur_size + extra_bytes) >> 2;
+ lrh0 = IPS_LRH_BTH;
+ if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
+ qp->s_hdrwords += ipath_make_grh(dev, &qp->s_hdr.u.l.grh,
+ &qp->remote_ah_attr.grh,
+ qp->s_hdrwords, nwords);
+ lrh0 = IPS_LRH_GRH;
+ }
+ lrh0 |= qp->remote_ah_attr.sl << 4;
+ qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
+ qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
+ qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords +
+ SIZE_OF_CRC);
+ qp->s_hdr.lrh[3] = cpu_to_be16(ipath_layer_get_lid(dev->dd));
+ bth0 |= ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
+ bth0 |= extra_bytes << 20;
+ ohdr->bth[0] = cpu_to_be32(bth0);
+ ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
+ ohdr->bth[2] = cpu_to_be32(bth2);
+
+ /* Check for more work to do. */
+ goto again;
+
+clear:
+ clear_bit(IPATH_S_BUSY, &qp->s_flags);
+bail:
+ return;
+}
OpenPOWER on IntegriCloud