/*
- * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
* SOFTWARE.
*/
+#include <linux/io.h>
+
#include "ipath_verbs.h"
#include "ipath_kernel.h"
* Note that we are in the responder's side of the QP context.
* Note the QP s_lock must be held.
*/
-static int ipath_make_rc_ack(struct ipath_qp *qp,
- struct ipath_other_headers *ohdr,
- u32 pmtu, u32 *bth0p, u32 *bth2p)
+static int ipath_make_rc_ack(struct ipath_ibdev *dev, struct ipath_qp *qp,
+ struct ipath_other_headers *ohdr, u32 pmtu)
{
struct ipath_ack_entry *e;
u32 hwords;
}
qp->s_hdrwords = hwords;
qp->s_cur_size = len;
- *bth0p = bth0 | (1 << 22); /* Set M bit */
- *bth2p = bth2;
+ ipath_make_ruc_header(dev, qp, ohdr, bth0, bth2);
return 1;
bail:
/**
* ipath_make_rc_req - construct a request packet (SEND, RDMA r/w, ATOMIC)
* @qp: a pointer to the QP
- * @ohdr: a pointer to the IB header being constructed
- * @pmtu: the path MTU
- * @bth0p: pointer to the BTH opcode word
- * @bth2p: pointer to the BTH PSN word
*
* Return 1 if constructed; otherwise, return 0.
- * Note the QP s_lock must be held and interrupts disabled.
*/
-int ipath_make_rc_req(struct ipath_qp *qp,
- struct ipath_other_headers *ohdr,
- u32 pmtu, u32 *bth0p, u32 *bth2p)
+int ipath_make_rc_req(struct ipath_qp *qp)
{
struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
+ struct ipath_other_headers *ohdr;
struct ipath_sge_state *ss;
struct ipath_swqe *wqe;
u32 hwords;
u32 len;
u32 bth0;
u32 bth2;
+ u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
char newreq;
+ unsigned long flags;
+ int ret = 0;
+
+ ohdr = &qp->s_hdr.u.oth;
+ if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+ ohdr = &qp->s_hdr.u.l.oth;
+
+ /*
+ * The lock is needed to synchronize between the sending tasklet,
+ * the receive interrupt handler, and timeout resends.
+ */
+ spin_lock_irqsave(&qp->s_lock, flags);
/* Sending responses has higher priority over sending requests. */
if ((qp->r_head_ack_queue != qp->s_tail_ack_queue ||
(qp->s_flags & IPATH_S_ACK_PENDING) ||
qp->s_ack_state != OP(ACKNOWLEDGE)) &&
- ipath_make_rc_ack(qp, ohdr, pmtu, bth0p, bth2p))
+ ipath_make_rc_ack(dev, qp, ohdr, pmtu))
goto done;
if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_SEND_OK) ||
qp->s_hdrwords = hwords;
qp->s_cur_sge = ss;
qp->s_cur_size = len;
- *bth0p = bth0 | (qp->s_state << 24);
- *bth2p = bth2;
+ ipath_make_ruc_header(dev, qp, ohdr, bth0 | (qp->s_state << 24), bth2);
done:
- return 1;
-
+ ret = 1;
bail:
- return 0;
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+ return ret;
}
/**
static void send_rc_ack(struct ipath_qp *qp)
{
struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
+ struct ipath_devdata *dd;
u16 lrh0;
u32 bth0;
u32 hwords;
+ u32 __iomem *piobuf;
struct ipath_ib_header hdr;
struct ipath_other_headers *ohdr;
unsigned long flags;
+ spin_lock_irqsave(&qp->s_lock, flags);
+
/* Don't send ACK or NAK if a RDMA read or atomic is pending. */
if (qp->r_head_ack_queue != qp->s_tail_ack_queue ||
(qp->s_flags & IPATH_S_ACK_PENDING) ||
qp->s_ack_state != OP(ACKNOWLEDGE))
goto queue_ack;
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+
+ dd = dev->dd;
+ piobuf = ipath_getpiobuf(dd, 0, NULL);
+ if (!piobuf) {
+ /*
+ * We are out of PIO buffers at the moment.
+ * Pass responsibility for sending the ACK to the
+ * send tasklet so that when a PIO buffer becomes
+ * available, the ACK is sent ahead of other outgoing
+ * packets.
+ */
+ spin_lock_irqsave(&qp->s_lock, flags);
+ goto queue_ack;
+ }
+
/* Construct the header. */
ohdr = &hdr.u.oth;
lrh0 = IPATH_LRH_BTH;
lrh0 = IPATH_LRH_GRH;
}
/* read pkey_index w/o lock (its atomic) */
- bth0 = ipath_get_pkey(dev->dd, qp->s_pkey_index) |
+ bth0 = ipath_get_pkey(dd, qp->s_pkey_index) |
(OP(ACKNOWLEDGE) << 24) | (1 << 22);
if (qp->r_nak_state)
ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPATH_MSN_MASK) |
hdr.lrh[0] = cpu_to_be16(lrh0);
hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
hdr.lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC);
- hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid);
+ hdr.lrh[3] = cpu_to_be16(dd->ipath_lid);
ohdr->bth[0] = cpu_to_be32(bth0);
ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & IPATH_PSN_MASK);
- /*
- * If we can send the ACK, clear the ACK state.
- */
- if (ipath_verbs_send(dev->dd, hwords, (u32 *) &hdr, 0, NULL) == 0) {
- dev->n_unicast_xmit++;
- goto done;
- }
+ writeq(hwords + 1, piobuf);
- /*
- * We are out of PIO buffers at the moment.
- * Pass responsibility for sending the ACK to the
- * send tasklet so that when a PIO buffer becomes
- * available, the ACK is sent ahead of other outgoing
- * packets.
- */
- dev->n_rc_qacks++;
+ if (dd->ipath_flags & IPATH_PIO_FLUSH_WC) {
+ u32 *hdrp = (u32 *) &hdr;
+
+ ipath_flush_wc();
+ __iowrite32_copy(piobuf + 2, hdrp, hwords - 1);
+ ipath_flush_wc();
+ __raw_writel(hdrp[hwords - 1], piobuf + hwords + 1);
+ } else
+ __iowrite32_copy(piobuf + 2, (u32 *) &hdr, hwords);
+
+ ipath_flush_wc();
+
+ dev->n_unicast_xmit++;
+ goto done;
queue_ack:
- spin_lock_irqsave(&qp->s_lock, flags);
+ dev->n_rc_qacks++;
qp->s_flags |= IPATH_S_ACK_PENDING;
qp->s_nak_state = qp->r_nak_state;
qp->s_ack_psn = qp->r_ack_psn;
wc->vendor_err = 0;
wc->byte_len = 0;
wc->qp = &qp->ibqp;
+ wc->imm_data = 0;
wc->src_qp = qp->remote_qpn;
+ wc->wc_flags = 0;
wc->pkey_index = 0;
wc->slid = qp->remote_ah_attr.dlid;
wc->sl = qp->remote_ah_attr.sl;
static inline void update_last_psn(struct ipath_qp *qp, u32 psn)
{
- if (qp->s_wait_credit) {
- qp->s_wait_credit = 0;
- tasklet_hi_schedule(&qp->s_task);
+ if (qp->s_last_psn != psn) {
+ qp->s_last_psn = psn;
+ if (qp->s_wait_credit) {
+ qp->s_wait_credit = 0;
+ tasklet_hi_schedule(&qp->s_task);
+ }
}
- qp->s_last_psn = psn;
}
/**
/* If this is a partial ACK, reset the retransmit timer. */
if (qp->s_last != qp->s_tail) {
spin_lock(&dev->pending_lock);
- list_add_tail(&qp->timerwait,
- &dev->pending[dev->pending_index]);
+ if (list_empty(&qp->timerwait))
+ list_add_tail(&qp->timerwait,
+ &dev->pending[dev->pending_index]);
spin_unlock(&dev->pending_lock);
/*
* If we get a partial ACK for a resent operation,
wc.vendor_err = 0;
wc.byte_len = 0;
wc.qp = &qp->ibqp;
+ wc.imm_data = 0;
wc.src_qp = qp->remote_qpn;
+ wc.wc_flags = 0;
wc.pkey_index = 0;
wc.slid = qp->remote_ah_attr.dlid;
wc.sl = qp->remote_ah_attr.sl;
list_move_tail(&qp->timerwait,
&dev->pending[dev->pending_index]);
spin_unlock(&dev->pending_lock);
+
+ if (opcode == OP(RDMA_READ_RESPONSE_MIDDLE))
+ qp->s_retry = qp->s_retry_cnt;
+
/*
* Update the RDMA receive state but do the copy w/o
* holding the locks and blocking interrupts.
qp->r_ack_psn = qp->r_psn - 1;
goto send_ack;
}
+ /*
+ * Try to send a simple ACK to work around a Mellanox bug
+ * which doesn't accept a RDMA read response or atomic
+ * response as an ACK for earlier SENDs or RDMA writes.
+ */
+ if (qp->r_head_ack_queue == qp->s_tail_ack_queue &&
+ !(qp->s_flags & IPATH_S_ACK_PENDING) &&
+ qp->s_ack_state == OP(ACKNOWLEDGE)) {
+ spin_unlock_irqrestore(&qp->s_lock, flags);
+ qp->r_nak_state = 0;
+ qp->r_ack_psn = qp->s_ack_queue[i].psn - 1;
+ goto send_ack;
+ }
/*
* Resend the RDMA read or atomic op which
* ACKs this duplicate request.
static void ipath_rc_error(struct ipath_qp *qp, enum ib_wc_status err)
{
unsigned long flags;
+ int lastwqe;
spin_lock_irqsave(&qp->s_lock, flags);
qp->state = IB_QPS_ERR;
- ipath_error_qp(qp, err);
+ lastwqe = ipath_error_qp(qp, err);
spin_unlock_irqrestore(&qp->s_lock, flags);
+
+ if (lastwqe) {
+ struct ib_event ev;
+
+ ev.device = qp->ibqp.device;
+ ev.element.qp = &qp->ibqp;
+ ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+ qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+ }
}
static inline void ipath_update_ack_queue(struct ipath_qp *qp, unsigned n)
case OP(SEND_FIRST):
if (!ipath_get_rwqe(qp, 0)) {
rnr_nak:
- /*
- * A RNR NAK will ACK earlier sends and RDMA writes.
- * Don't queue the NAK if a RDMA read or atomic
- * is pending though.
- */
- if (qp->r_nak_state)
- goto done;
qp->r_nak_state = IB_RNR_NAK | qp->r_min_rnr_timer;
qp->r_ack_psn = qp->r_psn;
goto send_ack;